Do Parents Tax Their Children? Teenage Labour Supply and ...
-
Upload
truongdien -
Category
Documents
-
view
219 -
download
1
Transcript of Do Parents Tax Their Children? Teenage Labour Supply and ...
Forschungsinstitut zur Zukunft der ArbeitInstitute for the Study of Labor
DI
SC
US
SI
ON
P
AP
ER
S
ER
IE
S
Do Parents Tax Their Children?Teenage Labour Supply and Financial Support
IZA DP No. 10040
July 2016
Angus Holford
Do Parents Tax Their Children?
Teenage Labour Supply and Financial Support
Angus Holford ISER, University of Essex
and IZA
Discussion Paper No. 10040 July 2016
IZA
P.O. Box 7240 53072 Bonn
Germany
Phone: +49-228-3894-0 Fax: +49-228-3894-180
E-mail: [email protected]
Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity. The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public. IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.
IZA Discussion Paper No. 10040 July 2016
ABSTRACT
Do Parents Tax Their Children? Teenage Labour Supply and Financial Support*
This paper models child employment and parental pocket money decisions as a non-cooperative game. Assuming that the child human capital is a household public good and that the relationship between child human capital and employment is concave, we compare the welfare obtained under different decision-making mechanisms and test the predictions of the model for a cohort of English teenagers in compulsory education. Our results support a situation in which parents ‘tax’ their children’s earnings, withdrawing financial support as the child increases his working hours. This strategy forces the child to internalise the social cost of his activities. JEL Classification: C52, C72, D13, J22 Keywords: intra-household transfers, pocket money, child labour supply,
noncooperative game, human capital Corresponding author: Angus Holford Institute for Social and Economic Research University of Essex Wivenhoe Park Colchester, CO4 3SQ United Kingdom E-mail: [email protected]
* This work was carried out during PhD studies supported by an Economic and Social Research Council ‘1+3’ studentship in Economics [reference number ES/I025499/1], and postdoctoral research supported by the Economic and Social Research Council Research Centre on Micro-Social Change [reference number ES/L009153/1]. Thanks in particular for advice from supervisors and colleagues Emilia Del Bono, Stephen Pudney and Matthias Parey, and my PhD examiners Pedro Carneiro and Thomas Crossley, as well as seminar participants at Essex, the Family Economics Workshop at Royal Holloway, and the IZA Summer School in Labour Economics. All errors are my own. This work is based on data from the Longitudinal Study of Young People in England (Secure Data Service Access), produced by the Department for Education and National Centre for Social Research, and supplied through the Secure Data Service at the UK Data Archive. This work also uses local labour market statistics produced by the Office for National Statistics. The data are crown copyright and reproduced with the permission of the controller of the HMSO and Queens’s Printer for Scotland. The use of the data in this work does not imply the endorsement of the Department for Education, National Centre for Social Research, ONS, or Secure Data Service at the UK Data Archive in relation to the interpretation or analysis of these data.
IZA Discussion Paper No. 10040 July 2016
NON-TECHNICAL SUMMARY
Young adults who are still living with their parents and still in compulsory education finance their independent consumption using either parental transfers (pocket money), or by working part-time. Parents care about their own and their children’s consumption, but also about the effect that part-time work might have on their children’s study time and academic achievement, so are prepared to sacrifice some of their income to subsidize their children’s expenses. This paper investigates how parents and their teenage children interact to set the amounts of (i) pocket money and (ii) part-time work. We use data on part-time working hours, and pocket money from parents, reported by around 5000 girls and 5000 boys in compulsory education in England, interviewed annually between 2004 (when they were 14) and 2006 (when they were 16). At age 14, 25% of boys but only 19% of girls had a part-time job. By age 16, girls had overtaken (32% against 29%) but were earning slightly lower wages for similar hours (£4.06 against £4.34 per hour, both for about 6.5 hours a week). The propensity to offer regular allowances from parents was very similar for boys and girls, at around 80% over this period. Children take into account the money they get from parents when deciding whether to work or not. At age 14 they are 10-15 percentage points less likely to work if they are receiving an allowance, and this gap increases to 20 percentage points by age 16. Similarly, parents are around 15 percentage points less likely to give an allowance if their child is in paid work. Our empirical analysis of parents’ and children’s behaviour reveals two key findings. The first is that, other things equal, parents are more likely to give regular pocket money to teenagers who will have a harder time finding a job. This includes teenagers who are younger within their academic year at school so at a disadvantage when vacancies arise in the autumn, both before the run-up to Christmas and when older cohorts leave for university; those living in areas with higher rates of adult unemployment; and those living in areas from which the nearest shops are less accessible. This tells us that rather than being opportunistic (“these parents don’t need to pay their children not to work, so they don’t”), parents effectively ‘insure’ their child’s independent consumption against labour market difficulties outside their control. They would like them to finance their own personal spending money, but will not penalize them if this proves difficult. The second is that parents are less likely give regular pocket money to teenagers, the longer the teenager’s hours of part-time work. In other words, parents effectively tax their children’s earnings from part-time work, by taking away cash they would otherwise be handing over. This effect is larger for girls than boys and strongest when they are 16 years-old. We suggest that close to the high-stakes GCSE exams, which determine these children’s future educational and labour market opportunities, parents become more inclined to use their financial resources to reduce their child’s incentive to work, hoping they will spend more time studying instead.
1 Introduction
As teenagers become older they increasingly demand independence over their consumption and
time use (Dauphin et al., 2011). However, while remaining in compulsory education they are
dependent on in-kind or financial transfers from parents to support their living expenses. They
may also undertake part-time work in the labour market. Many do. The data used here shows
that around 25% of 13-16 year-olds below the minimum school leaving age in England have a job
during school term time. In the US, employment rates are considerably higher: approximately
60% for 14-15 year-olds (Buscha et al., 2008, Bureau of Labor Statistics, 2000), with 92% of
the National Longitudinal Study of Youth (NLSY) 1997 cohort having worked at some point
during High School (Hotz et al., 2002).
Having a job while still at school may improve teenagers’ welfare by affording them additional
independent consumption, and potentially improves their stock of human capital in the form
of financial literacy, communication skills and lower discount rates (Oettinger, 1999; Light,
2001). Conversely, by crowding out time and effort devoted to education, in-school employment
may impair the child’s academic performance and subsequent educational or labour market
opportunities (Ruhm, 1997, Department for Children, Schools and Families, 2009). Above a
moderate number of hours per week the latter effect has been shown to dominate. This critical
threshold also becomes lower, the closer to exam time (Lillydahl, 1990; Payne, 2004).
Being unable to earn a living wage, teenagers in full-time compulsory education have no ‘outside
option’ or bargaining power within the household (McElroy and Horney, 1981; Browning and
Chiappori, 1998). In this environment, it is convenient to abstract from the child’s actions
and preferences when modelling the household’s decision-making processes. Several studies
either implicitly invoke Becker’s (1974, 1981) ‘rotten kid theorem’, assuming the existence of an
altruistic head-of-household who is able to transfer resources among household members and so
(under certain conditions - Bergstrom, 1989; Ermisch, 2003) automatically induce the child to
maximise household utility (Bingley and Walker, 2013), or treat children as goods rather than
agents (Bourguignon, 1999; Blundell et al., 2005). When teenagers are able to take part-time
employment these characterisations are less easily supported. The child’s labour market activity
increases the household’s full income, but also either contributes to or reduces the child’s level of
human capital, which in the presence of an altruistic parent can be treated as a household public
good. This situation is similar to the time allocation decision faced by a couple over labour
1
supply and household production, wherein the free rider problem leads to underprovision of the
public good (Konrad and Lommerud, 1995). Assuming the parent has a lower discount rate
than the child (Burton et al., 2002), making her paternalistic or impurely altrustic, she will
weight the child’s human capital more heavily relative to the child’s consumption than will the
child himself. The child’s labour market activity therefore creates an externality: the ‘rotten
kid theorem’ does not apply because in the absence of a specific mechanism design the child
will work too many hours and produce too little human capital, relative to the (parent’s) social
optimum.
In this paper we identify the decision-making process governing parent-child interactions in
respect of the child’s part-time labour supply. We consider three ways in which parents can
use their financial resources to alter their child’s incentive to get a job. They can set a fixed
transfer level taking the child’s decision as given (a Nash equilibrium), or announce their transfer
in advance of the child’s decision (a parent-leading Stackelberg model), or withdraw transfers
in response to the child’s employment (a child-leading Stackelberg model).
The former two models assume that children take their parents’ transfer level as given when
making their labour supply choice. This is the most common assumption so far adopted in the
literature. Dustmann et al. (2009), using data on UK 14-15 year-olds in 1974, show children’s
labour supply is highly responsive to parental transfers, with an income elasticity similar to that
of adults. Gong (2009) (with data on US 12-16 year-olds in 1997) and Kalenkoski and Pabilo-
nia (2010) (US college students between 1997 and 2003) found a significant but more modest
elasticity. Wolff (2006), analysed French young adults aged 16-22 in 1992 and found no effect.
As well as providing new evidence for a recent UK cohort (14-16 year olds in 2004-2006), we
extend this literature by explicitly considering the implications of an impurely altruistic parent
who cares relatively more about the child’s human capital, and by exploring the implications
of the child-leading Stackelberg model, in which parents set a contingent transfer schedule, as
a potential alternative decision making process. This model implies that parents effectively tax
their child’s earnings.
As noted by Kooreman and Kapteyn (1990), a problem for applied work on household labour
supply is that it is not clear which theoretical model to implement empirically, and Apps and
Rees (1996) show that estimating misspecified models of intra-household income and labour
supply interactions can result in highly misleading inference. We address this key identification
2
problem by inferring the decision-making process used by households in two stages. First we
analyse the welfare implications of each model in a theoretical framework, then we take testable
predictions from each model to the data. This approach contrasts with other work in which the
decision-making process for parent-child interactions has been elicited directly (Lundberg et al.,
2009), which carries the potential for responses to to be driven by social desirability (Conti and
Pudney, 2011) or for households’ arrangements to be tacit rather than explicit, making answers
unreliable.
In our first step we show that with an impurely altruistic parent the Nash equilibrium is Pareto-
dominated, and so should be rejected on welfare grounds: both the parent and the child will
always be better off at a Stackelberg equilibrium. However, the parent-leading and the child-
leading Stackelberg models are non-nested and neither is a priori welfare-superior, so we then
test predictions derived from these models using data from the Longitudinal Study of Young
People in England (LSYPE). The parent leading Stackelberg would predict that parents are
opportunistic in terms of reducing financial transfers in response to factors restricting the child’s
work opportunities, but we instead show that parents insure their child’s consumption against
labour market difficulties outside their control (Ermisch, 2003, p.5). Our results are consistent
with a decision-making process in which parents ‘tax’ their children’s earnings, withdrawing
financial support as child labour supply increases. This is in line with the predictions of the
child-leading Stackelberg model.
While there is strong evidence about the primacy of the role of parents in promoting financial
literacy and financial socialization of young adults, outweighing the contributions from part-time
work or from formal education (Shim et al., 2010), giving pocket money alone (characteristic of
the Nash and parent-leading Stackelberg models) is shown to be ineffective for stimulating adult
savings (Bucciol and Veronesi, 2014). Our finding shows that parents combine their giving of
pocket money with communication about a trade-off teenagers face when making an important
financial decision, which is an encouraging model for promoting the the financial literacy of
young adults.
We make an additional contribution to the empirical literature by repeating our analysis on the
same cross-section of parents and children in successive years. This enables us to evaluate how
the parameters of the model change as teenagers get older, while holding constant the other
factors (country, education level, era) collectively driving the differences between the labour
3
supply and transfer elasticities seen in the earlier literature. In particular, we show a significant
increase in the propensity of parents to ‘tax’ the part-time earnings of girls during their final
year of compulsory schooling when high stakes GCSE examinations are taken over earlier years.
This is consistent with parents using their financial resources to discourage part-time work,
during the periods when the opportunity cost of part-time work is highest.
2 Theoretical Model
Our model assumes two agents, a selfish child and an altruistic parent. The child cares about
his own consumption (C) and his own human capital (H). The parent cares about her own
consumption (P ), the child’s consumption, and the child’s human capital. The utility functions
depend on each agent’s relative preference for child’s human capital over child’s consumption,
denoted κc (child) and κp (parent). Here we assume that the parent’s altruism is impure or
paternalistic, meaning that κp > κc. Each agent’s κ is a predetermined function of exogenous
individual and household preferences and characteristics, X, so κi = κi(X) for i = p, c. The
utility functions can be written U c = U c(C,H, κc(X)) (child) and Up = Up(C,P,H, κp(X))
(parent). Both are three times continuously differentiable in C, P and H; strictly increasing in
C, P and H at a dimishing rate; and separable in C, P and H, such that the marginal utility
from each element is independent of the other elements.1
Prior to making his labour supply decision, the child has a baseline endowment of human capital,
µ. This is also a predetermined function of X, so µ = µ(X). These current and past parental,
child and household characteristics are assumed to have determined the parent and child’s time
and effort devoted to investment activities over the preceding lifecourse. The child’s human
capital accumulation in a given period is also a function of his time devoted to wage-earning
employment L, so H = H(L, µ). We assume that the marginal human capital product (HL)
of the first hour of labour supply can be positive or negative, but (HL) is decreasing in L
(so HLL < 0). This captures the idea that beyond some level of hours of work employment
must detract from the quality or quantity of time devoted to other human-capital earning
pursuits. This is consistent with the findings of Ruhm (1997), Tyler (2003) and Montmarquette
et al. (2007), among others, and we assume that this stylised fact is built into individuals’
1Formally, the restrictions on the child’s utility function are: UcC > 0; Uc
CC < 0; UcH > 0; Uc
HH < 0; UcCH = 0.
The restrictions on the parent’s utility function are: UpC > 0; Up
CC < 0; UpP > 0; Up
PP < 0; UpH > 0; Up
HH < 0;U
pPH = 0; Up
PC = 0; UpHC = 0.
4
expectations.
In our model, the parent has exogenous income M , and can choose to transfer the cash amount
t ≥ 0 to the child. The parent consumes the rest of her income, so P = M − t. The child
chooses his labour supply, L, and earns the constant wage w > 0. His consumption is the sum
of labour earnings wL, and money received from parents: C = wL+ t.
Observed transfer and labour supply behaviour represents the solution to the one-shot game
in which the parent and child each maximise their own utility in the current period subject to
their preferences and production function of human capital.
We assume that parents and children do not have a mechanism for a complete, binding commit-
ment. This means we adopt a non-cooperative model. This is supported by empirical evidence
showing there is imperfect communication between the parent and child during adolescence, a
period in which young people establish their identity and develop the ability to make their own
decisions (Collins, 1990). As Lundberg et al. (2009) put it, “effective information exchange
[is] hampered by the [child’s] still-developing cognitive and communicative abilities”. These
communication difficulties stem from both the asymmetric relationship between parents and
adolescents and divergent expectations about the child’s autonomy (Steinberg, 2001; Collins
and Luebker, 1994), and from the changing nature of the relationship over time (Collins et
al, 1997), leading to an unstable bargaining environment (“constantly changing as the child
matures”; Lundberg et al., 2009).
In our theoretical model, we follow Burton et al. (2002) and Schmidt (1993) by assuming that
the parent has developed a reputation and the child’s discount rate is sufficiently high that
he takes the parent’s strategy (though not necessarily action) as given in each period. We
then build on two subsequent approaches to modelling household labour supply adopted in the
earlier literature (see Apps and Rees, 2009, pp.75-76, for a discussion). Firstly we model the
child’s consumption and human capital as a household public good, and secondly we derive and
identify best response functions for both agents’ decision variables (Antman, 2012; Leuthold,
1968; Ashworth and Ulph, 1981).
In the following sections will derive the unique equilibrium under (i) the Nash decision-making
process, (ii) the parent-leading Stackelberg model, and (iii) child-leading Stackelberg model.
This final model can also be framed as the parent setting a contingent transfer schedule.
5
2.1 Nash and Stackelberg equilibria
In a Nash framework, the parent and child choose their cash transfer and labour supply as
their best response to the other’s action. At the Nash equilibrium, neither agent can do better,
taking the other’s action as given. Implicitly, this position is reached by a myopic adjustment
procedure or iterative process in which the agents choose starting positions and best-respond
in turn to converge on the equilibrium (Leonard, 1994). In the Stackelberg framework, one
party ‘leads’ by committing to the action that maximises their own utility taking into account
that the other agent will ‘best respond’ to the leader’s actions. Here, there are two Stackelberg
equilibria, one in which the parent leads and one in which the child leads.
In the parent-leading Stackelberg model, the parent fixes the cash transfer in advance. The
child follows with his best response in terms of employment, taking the cash transfer as given.
We assume the parent can costlessly and rapidly alter the transfers. However, this flexibility
threatens the parent’s credibility or time-consistency. A Stackelberg equilibrium requires the
follower to believe the leader will not deviate from her initial action. In an application of the
‘Samaritan’s Dilemma’ (Bruce and Waldman, 1991; Lindbeck and Weibull, 1988) this means
that the parent must hold the transfer level fixed in case of the child being involuntarily un-
employed or under-employed, and that parental transfers do not respond to changes in labour
market conditions.
In the child-leading Stackelberg model, the child anticipates the parent’s best response and
‘leads’ by choosing his employment level. The parent then follows with his best response in terms
of cash transfers, taking the employment level as given. We assume the parent has deterministic
control over cash transfers. This means the child’s first-move commitment will not come under
challenge due to the parent deviating from her expected best-response for reasons outside her
control. The child’s labour supply decision is less reversible than the parent’s transfer decision.
The child can leave a job but not easily find another job or adjust his hours. These considerations
mean the employment level is more likely to represent a credible, binding commitment than the
transfer level.
The parent’s relative financial resources, lower discount rate, and position of authority within
the household, and the child’s lack of an outside option should all mean the parent has all of the
bargaining power (McElroy and Horney, 1981; Browning and Chiappori, 1998). The idea of a
child-leading Stackelberg model, in which the parent ‘does the best she can given the action of
6
a dominant child’ is therefore not intuitively appealing. However, this scenario can be reframed
as the parent announcing a contingent schedule of transfers that will be made for every possible
level of employment the child can take, and the child choosing the point along it which is best
for him. Note that with a fully contingent strategy, there is no time-consistency problem for
the parent.
Hence, both the parent-leading and child-leading Stackelberg models are consistent with a
situation in which the parent has the bargaining power and chooses the decision-making process.
We assume she chooses that which her best off (Bell et al., 2012). We now evaluate the welfare
outcomes of the three possible models.
2.2 Comparing the outcomes of alternative models
Each agents’ optimisation problem can be illustrated graphically in a utility map in the [transfer,
labour supply] space, as shown in Figure 1.
Here, the child has a ‘bliss point’ (point of maximum utility) at infinite transfers and the
human capital maximising level of labour supply. The child’s best response correspondence,
BRc = L∗(t, w,X), is the locus of points maximising the child’s utility for each given transfer
level. This passes through the turning point of each indifference curve (U c1 , U
c2 , etc). Above
the BRc, the indifference curve is upward sloping: the marginal human capital cost of working
is so high that he requires additional transfers from the parent as well as his own earnings to
compensate for it.
As the parent cares relatively more about human capital than does the child, the parent’s ‘bliss
point’ lies below the child’s best-response correspondence. If the parent is very selfish, the high
cost in terms of her own consumption of increasing the child’s human capital and consumption
will result in a bliss point below the human-capital maximising level of labour supply. The
parent’s best response correspondence BRp = t∗(L,M,X) is the locus of points maximising the
parent’s utility for each given labour supply. This passes through the maximum and minimum
points of her concentric indifference curves (Up1 , U
p2 , etc). These carry decreasing utility coming
from suboptimal consumption allocations when increasing or decreasing transfers either side of
the bliss point, and reduced human capital when increasing labour supply above, or reducing
below, the human capital maximising level.
Although next we will set out each agents’ optimality conditions and the resulting welfare
7
Figure 1: Agents’ optimisation problem and the three equilibria
Parent
bliss
point Human
cap’ max’
0 → t t
L
↑
0
L
= t*(L, M, X)
= L*(t, w, X)
Child bliss
point at
t=∞
Nash equilibrium
Child-leading
Stackelberg
equilibrium
Parent-leading
Stackelberg
equilibrium
Increasing child’s utility
Increasing parent’s utility
comparisons formally, this utility map shows the intuition for a straightforward interior solution
case. The Nash equilibrium, or mutual best response, is at the intersection of the agents’ best-
response correspondences. Both Stackelberg equilibria entail the leader choosing the point on
the follower’s BR correspondence which makes him or her best off.2
2.2.1 Nash versus parent-leading Stackelberg
The parent’s maximisation problem is the same under a Nash equilibrium as when she acts as
a Stackelberg follower. In either case, when choosing her transfer level she takes the child’s
labour supply L as given:
maxt
Up = Up(P (M − t), C(wL+ t), H(L, µ(X), κp(X)) (1)
2Figure 1 shows an interior solution. Because transfers must be non-negative, if at t = 0 the parent’s marginalutility from her own consumption exceeds that from the child’s human capital and consumption, she will make zerotransfers. If the human capital maximising level of labour supply is L = 0 and for the first hour of employmentthe child’s marginal utility lost from human capital is greater than that gained from increased consumption, thechild will work zero hours.
8
and her optimality condition is:
UpC = U
pP (2)
This states that the parent will make transfers to the point where her marginal utility from (i)
her own consumption, and (ii) the child’s consumption and human capital, are equal. This will
occur at a maximum or minimum point of a parental indifference curve.
Acting as a Stackelberg leader, the parent’s maximisation problem must account for the indirect
effects of t on her utility from the child’s consumption and human capital, caused by the child’s
labour supply response to t:
maxt
Up = Up(P (M − t), C(wL∗(t) + t), H(L∗(t), µ(X)), κp(X)) (3)
Her optimality condition is now:
UpC +
∂L∗
∂t[Up
C · w + UpH ·HL] = U
pP (4)
She still makes transfers to the point where the marginal utilities from her own consumption
and the child’s consumption and human capital are equal, but this time accounting for the
child’s optimal labour supply response. This means she chooses the point on the child’s BR
correspondence that is tangent to a parental indifference curve. This will represent the parent’s
highest attainable utility under this framework.
The expression in square brackets in equation (4) is similar to the child’s Stackelberg-leading
optimality condition (equation 9, below): w.U cC = −HL.U
cH , or equivalently w·U c
C+HL ·UcH = 0.
Both expressions are equal to the net change in the parent’s utility caused by simultaneous
changes in the child’s consumption and human capital. If the parent were purely altruistic,
placing the same relative weight on the child’s human capital and consumption as does the
child, then [UpC · w + U
pH ·HL] and [U c
C · w + U cH ·HL] would both equal zero at the same level
of labour supply and transfers. This would mean the parents’ bliss point lies on the child’s BR,
and the Nash and parent-leading Stackelberg equilibria would coincide, as shown by Dustmann
et al. (2009). As we assume that the parent’s altruism is paternalistic however, at the Nash
9
level of transfers it holds that UpC · w + U
pH ·HL < 0, and hence:
UpC +
∂L∗
∂t[Up
C · w + UpH ·HL] > U
pP (5)
From this position, in the Stackelberg model, the parent’s paternalism makes her prepared to
pay the child more (increase t) to persuade him to work less (reduce L).3 It can be shown that
the child’s BR is sufficiently shallow for him always to withdraw labour earnings at a slower rate
than transfers are received (see Appendix A.3.1, page 53, for the complete derivation) and both
equilibria must lie in the range where the marginal human capital product of labour is negative
(HL < 0). Otherwise, the child and parent would always prefer to increase labour supply at
the benefit to both human capital and consumption. Hence, both the child’s consumption and
human capital must be higher at the parent-leading Stackelberg equilibrium than at the Nash
equilibrium. Moving from the Nash to the parent-leading Stackelberg equilibrium makes both
the parent and the child better off.
2.2.2 Nash versus child-leading Stackelberg
The child’s maximisation problem when choosing his labour supply is the same under a Nash
equilibrium as when he acts as a Stackelberg follower. In both cases, he takes the transfer, t,
made by the parent, as given:
maxL
U c = U c(C(wL+ t), H(L, µ(X)), κc(X)) (6)
and his optimality condition is:
w.U cC = −HL.U
cH (7)
meaning his marginal utility from additional consumption (his gross wage) is equal to that from
foregone human capital.
Acting as a Stackelberg leader the child’s optimisation problem is as follows:
maxL
U c = U c(C(wL+ t∗(L)), H(L), µ(X), κc(X)) (8)
3In Appendix A.1 (page 44) we consider the implications of a parent with a different kind of impure altruisticpreferences: those who derive no utility from the child’s human capital.
10
and his optimality condition is:
(w +∂t∗
∂L) · U c
C = −HL · U cH (9)
The parent withdraws transfers in response to increasing labour supply, so ∂t∗
∂L< 0, meaning
that when acting as a Stackelberg leader, the child instead bases his labour supply on his net
wage (w + ∂t∗
∂L) rather than gross wage. He selects the point on his parents’ best response
correspondence, which he can view as a contingent tax or transfer schedule, which gives him
highest utility. This will occur at a tangent point to a child indifference curve. Formally, at the
Nash level of labour supply it holds that:
(w +∂t∗
∂L) · U c
C < −HL · U cH (10)
At this position, the net wage is too small to compensate the child for the marginal human
capital cost of employment. He will therefore reduce his labour supply, raising his human
capital and reducing consumption, but become better off overall.
At equilibrium, the ‘tax rate’ - the slope of the best-response correspondence along which
cash transfers from parents are observed to be withdrawn in response to the child’s earnings - is
strictly between zero and one (see Appendix A.3.2, page 54, for the complete derivation). Above
the human capital maximising labour supply, the child will never work for a negative or zero net
wage, and the equilibrium cannot occur at a point where the parent is subsidizing the child’s
employment. The parent only does this where (i) the child’s human capital is still increasing
in labour supply or (ii) where she would prefer to increase the child’s consumption at the cost
of reducing her own consumption and her child’s human capital. The parent’s ‘bliss point’ will
occur at the point where the parent is indifferent between increasing the child’s consumption,
reducing the child’s human capital, and reducing her own consumption. In this state she
will neither tax nor subsidise the child’s employment. Hence, the child-leading Stackelberg
equilibrium must lie at a point on the parent’s best-response correspondence between the Nash
equilibrium and the parent’s bliss point. This means that both the child and the parent are
better off at the child-leading Stackelberg equilibrium than the Nash equilibrium.4
4It is a sufficient condition for this welfare improvement that the parent’s preferences are paternalistic. Onemay also consider that if the child-leading Stackelberg equilibrium lay below the parent’s bliss point, it wouldimply that while the child is happy with the trade-off between human capital and consumption, the parent wouldprefer to raise the child’s consumption and reduce his human capital. This would violate our assumption that
11
2.3 The identification problem
If the chosen decision-making process were known, the strategy for identifying each agent’s
BR correspondence can easily be visualised in Figure 1. In the Nash framework, each agent’s
BR correspondence is traced out by shifts in the other’s BR correspondence, caused by factors
having no direct effect on their own. In each Stackelberg model, shifts in the leader’s utility
map caused by factors having no direct effect on the follower’s BR will trace out this BR as the
locus of points which are tangent to the leader’s highest attainable indifference curve.
However, the chosen decision-making process is ex ante unknown, and estimating misspecified
models of intra-household income and labour supply interactions has been shown to result in
highly misleading inference (Apps and Rees, 1996). Following Apps and Rees’ approach, our
identification problem can be illustrated by considering the result we would obtain were we to
proceed under the belief that the Nash framework was the correct specification for household
behaviour when the true data generating process the is parent-leading Stackelberg model. Then,
shifting the child’s BR correspondence will trace out a locus of points which are tangent to the
parent’s indifference curves. This will incorrectly be interpreted as representing the parent’s BR
correspondence. The case shown in Figure 2 would lead us to conclude that at the equilibrium
margin parents are choosing to subsidise, rather than tax, their child’s earnings.
To identify the decision-making process we follow first a theoretical and then an empirical
argument. We can rule out the observed outcomes ever representing the Nash equilibrium
on welfare grounds. We showed in section 2.2, and it is apparent in Figure 1, that both the
parent and the child are better off at both the parent-leading and the child-leading Stackelberg
equilibria, meaning the Nash equilibrium is Pareto dominated. The parent-leading model is
superior even for the child because the parent’s paternalism makes him prepared to pay the
child more to work less. The child-leading model is superior even for the parent because the
positive marginal tax rate faced by the child means he internalises the social cost (in terms of
human capital) of his employment. This makes him reduce his labour supply, which pleases the
parent.
We assume the parent chooses the decision-making process which gives her the highest utility.
Our theoretical framework cannot indicate whether this will be the parent-leading or child-
leading Stackelberg model. Figure 1 is drawn assuming that the parent is indifferent between
the parent cares relatively more about human capital than does the child.
12
Figure 2: Erroneous identification of the parent’s best response.
Parent
bliss
point
0 → t t
L
↑
0
L
Shifts in child’s
best response
= L*(t, w, X)
Child bliss
point at
t=∞
True parent’s
best response Estimated
parent’s best
response
Observed data
points are
Stackelberg
equilibria
= t*(L, M, X)
Human
cap’ max’
these decision-making processes, receiving utility level UP2 in both cases. The more paternal-
istic are the parent’s preferences (characterised by a steeper subsidy and tax on employment
below and above the parent’s bliss point respectively), the greater scope there is to favour the
child-leading model.5 The more selfish is the parent overall (characteristed by shifting her BR
correspondence to the left and her bliss point to a lower level of labour supply, and reduc-
ing the tax and subsidy rates over the relevant ranges), the more scope there is to favour the
parent-leading model.
Since we cannot reject either Stackelberg model on theoretical grounds, we must do so on empir-
ical grounds. We now discuss testable predictions derived from the parent-leading Stackelberg
model that allow us to validate or reject this model.
2.4 Labour market rationing and the Stackelberg model
The parent-leading Stackelberg model requires that the parent will hold transfers constant
in response to factors expected to restrict the child’s optimal choice (his labour supply or
5In Appendix A.1 we derive the predictions of the model for the extreme case in which the parent gains noutility from the child’s human capital. In this case, we show that the parent-leading model is strictly preferredto the Nash, in turn strictly preferred to the child-leading model.
13
probability of obtaining a job). Labour demand conditions should have no effect on the agents’
preferences or utility except via the child’s labour supply. Therefore, a time-consistent parent
should ignore these factors in setting his transfer decision.
In addition, if a binding ceiling is placed on the child’s hours of employment, then the parent
should reduce the transfers made to the child: she no longer needs to ‘pay the child not to
work’ and at this ceiling can reduce transfers without trading her rise in consumption against
a reduction in the child’s human capital. Such a ceiling may result from legal restrictions on
the number and timing of hours of employment a child may offer, or from labour demand
constraints.
If the parent-leading Stackelberg model is true, therefore, we should find that the effect of
the labour demand variables expected to restrict the child’s employment opportunities on the
transfer are either zero or negative. If this prediction is rejected empirically, the only decision-
making process left standing is the child-leading Stackelberg model, and we present results
which are fully consistent with this framework being used.
3 Data and institutional background
3.1 LSYPE data
We test the predictions of our model using data from the first three waves of the Longitudinal
Study of Young People in England (LSYPE). This sample is drawn from a single academic
cohort of teenagers in England, interviewed annually from 2004-2006 at age 14 (school year 9),
15 (year 10), and 16 (year 11), the last three years of compulsory schooling. The data enable us
to control for a rich set of individual and household variables and the child’s prior educational
performance.
The question on parental cash transfers is “Do you receive any pocket money, allowances, or
other support towards your living costs from parents or relatives?”. The emphasis on “al-
lowances” or “support towards your living costs” but without reference to specific items means
that this does not refer to hypothecated transfers. The present tense makes clear that this
unearned income refers to a current, ongoing, and regular arrangement that represents a close
substitute for labour income. Since we observe changes in parental transfers only at the ex-
tensive margin, switching from positive to zero transfers, we assume that this variable reflects
14
parents’ latent desired transfer level, and that this changes monotonically with parental income,
child labour supply, and child’s labour market restrictions.
Youth respondents are also asked “Do you ever do any work in a spare-time paid job, even
if it is only for an hour or two now and then? (Please don’t include jobs you only do during
the school holidays or voluntary work)”. Those answering ‘yes’ are asked “How many hours on
average do you usually work in this job (or jobs) during a term time week? Please include any
hours you work at the weekend during term-time”. These questions explicitly exclude unpaid
or voluntary work, ensuring that a positive response is unambiguously associated with labour
income. The term-time focus makes the crowding out of study time or other extra-curricular
activities a salient concern.
3.2 Employment of children in England
The rules governing the employment of children in England are set by the Department for
Education (see guidelines in DCSF, 2009). Children aged at least 13 but less than the school
leaving age may undertake ‘light’ work, deemed as not being harmful to their health, safety or
development.6 There are age-specific restrictions on the types and hours of work children may
do. Those under 16 cannot work ‘mainly or solely’ for the sale of alcohol, for example. Those
in compulsory education may work only 12 hours per week in term time, including a maximum
of 2 hours on a weekday or Sunday; 8 hours on a Saturday (5 hours if under 15 years); one hour
before school on a weekday; and none during school hours or after 7pm on a school night. Those
employing children below the school leaving age must obtain the signature of a parent and a
permit from the Local Education Authority, which must be satisfied that the child’s education
will not be damaged. Howieson et al. (2006) estimate that in Scotland only 11% of children
covered by this legislation had the required permit, so these regulations are poorly enforced.
3.3 Descriptive statistics
Table 1 shows descriptive statistics relating to employment, earnings and cash transfers from
parents. The propensity to work rises as the children age, particularly among girls, who start
from a lower base. Conditional on working, girls work longer hours than boys, but receive lower
6Full-time education is compulsory until the last Friday in June of the academic year when the child turns 16.
15
earnings and hourly wages. The proportion receiving cash transfers declines very slightly with
age, and is marginally higher for girls than boys.
Table 1: Descriptive statistics: Employment and transfer receipt by gender
Wave 1 (Age 14) Wave 2 (Age 15) Wave 3 (Age 16)
Boys Girls Boys Girls Boys Girls
Employed (%) 24.9 18.9 28.3 27.1 29.0 31.2
Receive cash transfer (%) 80.9 81.7 77.8 79.4 76.7 78.4
Mean hours employment2 4.14 4.24 5.16 5.38 6.44 6.72(conditional on being employed) (0.10) (0.11) (0.13) (0.10) (0.15) (0.13)
Mean earnings2, £s 14.52 14.24 20.58 18.90 27.96 27.31(conditional on being employed) (0.41) (0.39) (0.52) (0.33) (0.63) (0.55)
Observations 7116 7250 6258 6303 5815 5839
Notes: All figures for hours and earnings are weekly. Standard errors in parentheses. Population means and proportions calculated usingfinal probability weights. Standard errors clustered by school. 1: GCSE capped point score (calculated over best eight subjects). 2: Hours
of employment and earnings are per week. 3: Key Stage 3 Average Point Score, standardized by subtracting mean and dividing by standarddeviation.
Table 2 shows the differences in prior educational performance (age 14) and subsequent ed-
ucational performance (age 16) according to the child’s employment status in each year. At
every stage, and for both boys and girls, those in employment are positively selected on age
14 educational performance, and go on to perform better at age 16 by a sufficient margin to
widen future educational opportunities substantially.7 This relationship is driven by the posi-
tive correlation between factors determining employability and academic performance, such as
individual motivation, relative age (Plug, 2001; Crawford et al., 2013), school quality, parental
investments, and local labour market conditions.
Table 2 also shows that the gender wage gap is not driven by differential selection: Girls
have superior prior educational performance and are more positively selected into employment
according to these grades. Instead, these observations are consistent with boys and girls being
active in distinct labour markets with different wage rates and factors restricting employment:
Kooreman (2009) shows the in-school employment gender wage differential in the Netherlands
is due to the concentration of girls in babysitting roles, for example. For this reason we estimate
7The difference in performance is equivalent to one GCSE grade in four of the student’s best eight subjects.The threshold for continuation in full-time education is five A*-C grades. This is achieved by around 60% ofstudents. Students typically take around 10 GCSE courses.
16
Table 2: Descriptive statistics: Prior and subsequent educational performance by employmentstatus
..Wave 1 (Age 14).. ..Wave 2 (Age 15).. ..Wave 3 (Age 16)..
Boys Girls Boys Girls Boys Girls
Standardized age 14 exam score3 by employment status
If employed 0.097 0.293 0.096 0.302 0.114 0.302(0.032) (0.034) (0.033) (0.031) (0.031) (0.034)
If not employed.......... -0.101 0.019 -0.090 -0.009 -0.094 -0.015(0.034) (0.030) (0.036) (0.018) (0.037) (0.034)
Mean age 16 exam score3 by employment status
If employed 294.31 328.06 295.07 331.37 299.44 334.50(3.35) (3.42) (3.47) (2.83) (2.83) (2.96)
If not employed.......... 276.75 303.62 277.41 301.36 276.39 300.91(3.50) (2.89) (3.57) (1.92) (3.69) (3.30)
Observations 7116 7250 6258 6303 5815 5839
Notes: All figures for hours and earnings are weekly. Standard errors in parentheses. Population means and proportions calculated usingfinal probability weights. Standard errors clustered by school. 1: GCSE capped point score (calculated over best eight subjects). 2: Hours
of employment and earnings are per week. 3: Key Stage 3 Average Point Score, standardized by subtracting mean and dividing by standarddeviation.
our models separately for males and females.8
Finally, Table 3 shows that employment and financial support are always negatively correlated.
Parents are always more likely to make cash transfers to children who do not work, and chil-
dren are always more likely to work if they do not receive cash transfers. This is consistent
with children’s working decision being financially motivated and with parents’ use of financial
transfers to disincentivise child employment.
4 Estimation
We estimate cross-sectional models of the observed counterparts to the parent-leading and child-
leading Stackelberg models. The econometric specification for the transfer and labour supply
decisions are shown in equations (11-12) and (13-14). In this exposition the functions and error
terms with superscript S refer to an agent acting as a Stackelberg leader. The superscript N
refers to an agent taking the actions of the other as given, as in a Nash equilibrium or acting
8In addition to social norms about the types of jobs to be taken by boys and girls, there are rigid genderroles within certain ethnic groups which make employment among girls exceptionally rare, both in comparisonto other ethnic groups and to boys of the same ethnic group. Appendix A.4, page 56, presents some descriptivestatistics on both these topics.
17
Table 3: Descriptive statistics: Negative correlation between child’s part-time employment andparental transfers
.Wave 1 (Age 14). .Wave 2 (Age 15). .Wave 3 (Age 16).
Boys Girls Boys Girls Boys Girls
Transfer receipt, %, by employment status:
If employed 71.5 73.3 66.3 71.1 63.2 67.5
If not employed 83.9 83.6 82.4 82.6 82.0 83.3
Employment rates, %, by transfer receipt
If receives transfers 22.0 17.0 24.2 24.3 23.1 26.8
If does not receive transfers..... 37.1 27.6 43.1 38.2 44.4 46.9
Observations 7116 7250 6258 6303 5815 5839
Notes: All figures for hours and earnings are weekly. Standard errors in parentheses. Population means and proportions calculated usingfinal probability weights. Standard errors clustered by school. 1: GCSE capped point score (calculated over best eight subjects). 2: Hours
of employment and earnings are per week. 3: Key Stage 3 Average Point Score, standardized by subtracting mean and dividing by standarddeviation.
as a Stackelberg follower.
The decision to estimate separately by wave rather than exploiting the panel nature of this
dataset is made because the teenage years are a time of transition. Teenagers have very different
labour market opportunities at age 16 than 13. Time away from studying may be increasingly
costly in terms of educational performance the closer the children get to the exams which will
determine their future academic opportunities. Children will desire higher consumption and
increasing independence over what they consume. For these reasons, we do not expect a stable
relationship over time that conventional panel data methods are appropriate for.9 Moreover,
this approach enables us to identify the effect of changing one contextual factor - the age of the
children - on the parameters of the model. This is a key contribution to a literature in which
several factors may be responsible for differences in previous findings.
9We use longitudinal weights to ensure that the estimates are obtained from the same sample of individualsat each wave and that the sample is representative of the population in terms of household demographics andparental background as measured at wave 1. See James et al. (2010) for details of the weights’ construction.
18
4.1 Parent-leading Stackelberg model
Equations (11-12) show the parent-leading Stackelberg model. Equation (11) represents the
parent’s transfer decision. Equation (12) represents the child’s effective best correspondence.
t = 1 if tS(M,Z,X) + ǫSt > 0 (11)
t = 0 if tS(M,Z,X) + ǫSt ≤ 0
L = LN (tS ,Z,X) + ǫNL if LN (tS ,Z,X) + ǫNL > 0 (12)
L = 0 if LN (tS ,Z,X) + ǫNL ≤ 0
Transfers and labour supply are subject to correlated random shocks due to unobservable het-
erogeneity in preferences, resources, and constraints such as discrete job offers and variation in
travel time to school or work, such that:
(
ǫStǫNL
)
∼ N
(
00
)
,
(
σSt ρ1ρ1 σN
L
)
This system of a simultaneous probit and tobit is estimated using Full Information Maximum
Likelihood (FIML). This system is identified through the exclusion of parental income, M
from the second stage (child’s labour supply) equation. In other words, we assume that the
probability of receiving a transfer is strictly increasing in parental income, but there is no direct
effect of parent income on child’s labour supply (Dustmann et al., 2009; Wolff, 2006; Kalenkoski
and Pabilonia, 2010; and Gong, 2009).10
Variables expected to restrict the child’s labour market opportunities, Z are included in both
equations to enable us to test whether (i) these variables have the predicted effect on child
labour supply, and (ii) the parent holds constant or reduces transfers in response to these
labour demand conditions.
10The LSYPE uses a different measure of income in each sampling wave. In waves 1 and 3 the variable“Total income from work, benefits and anything else for [Main Parent] (and partner)” is elicited in 92 bandsand 12 bands respectively. In wave 2, the “gross annual salary” of both parents and the “total annual amountof benefits received” are elicited separately, and other sources of income are omitted. Because these measuresare not directly comparable, we use a ‘relative permanent income’ measure defined as the percentile (zero beingthe lowest, one being the highest) of the household’s mean gross income percentile across the three waves, usingimputed household income where necessary. Details of the imputation procedure can be found in James et al.(2010).
19
4.2 Child-leading Stackelberg model
Equations (13-14) show the child-leading Stackelberg model. Equation (13) represents the
child’s labour supply decision, with the child accounting for the parent’s transfer schedule.
Equation (14) represents the parent’s best response, or contingent transfer schedule:
L = LS(M,Z,X) + ǫSL if LS(M,Z,X) + ǫSL > 0 (13)
L = 0 if LS(M,Z,X) + ǫSL ≤ 0
t = 1 if tN (L,M,X) + ǫNt > 0 (14)
t = 0 if tN (L,M,X) + ǫNt ≤ 0
We again assume the presence of correlated random shocks such that:
(
ǫNtǫSL
)
∼ N
(
00
)
,
(
σNt ρ2ρ2 σS
L
)
This system is identified through the exclusion of labour market factors Z from the second-stage
(parental transfer) equation.
We use three different ways to represent restrictions on the child’s labour supply. First, the
child’s relative age within the academic cohort is included as a linear term, equal to 1 for
September birthdays (the oldest in the cohort) up to 12 for those born the following August.
Children born earlier in the academic year will be allowed to work longer hours or access specific
job types earlier than their younger peers. This means they are better placed to fill suitable
vacancies which arise in the autumn as temporary staff are taken on prior to Christmas, and
members of older cohorts move away to university. The parent’s attitude to financial support
may however change when the child turns a year older, so we also control for the child being
interviewed after their birthday within the academic year.
Secondly, we use the 2004 index of multiple deprivation (IMD) of the teenagers’ area of residence
as a proxy for very localized area characteristics determining a teenager’s employment oppor-
tunities. The IMD measures the prevalence of deprivation within geographical areas containing
400-1200 households (Office of the Deputy Prime Minister, 2004). The index is calculated from
20
aggregated individual-level deprivation indicators and area characteristics in seven domains:
income; employment; health; education, skills and training; barriers to housing and services;
crime; and the living environment. All these elements are expected to be correlated with youth
labour market opportunities. Of greatest relevance in driving this relationship are (i) the area
unemployment rate, and (ii) the average road distance to a post office and supermarket or con-
venience store. The latter are suitable potential employers and may also be clustered with other
shops and businesses. The IMD is functionally dependent on the characteristics of the individual
households observed in the LSYPE. To net out this effect we explicitly control for household-
level counterparts to the componenents of the aggregated deprivation indicators recorded in the
IMD. We assume that conditional on household circumstances (including parental employment
and income and the child’s prior educational performance) the local area characteristics cap-
tured by the IMD have no direct effect on the parent’s transfer decision except via the child’s
employment opportunities.
Thirdly, we use the adult (aged 16-64) unemployment rate in the youth’s local authority district
(LAD) of residence at time of interview. There are 325 LADs in England, with an average
population of 164,000. This variable therefore represents labour market opportunities over a
wider geographic area. Conditional on the parents’ own income and employment status we
assume this has no direct effect on parent’s transfer decision, except via the child’s employment
opportunities.
4.3 Additional controls
The vector of additional controls, X, includes further variables expected to affect the parent’s
resource constraints, either agent’s relative preference for human capital (κi) or the child’s
labour market opportunities or endowment of human capital (µ). These comprise (i) socio-
economic characteristics, including parent’s employment status and education; (ii) the child’s
prior educational performance and special educational needs (SEN) status; (iii) ethnicity dum-
mies, to account for differing preferences and labour market discrimination; and (iv) household
structure, which determines individual’ material and time resources as well as preferences. We
account for geographical and seasonal variation in labour market opportunities with dummies
for urban-rural status, residence outside Greater London, and year and month of interview.
21
5 Results
In this section, we first present the estimates obtained under the parent-leading Stackelberg
framework. We show these are contrary to the predictions of the model in two ways. We then
show results consistent with the child-leading Stackelberg mechanism being used instead. The
complete maximum likelihood estimation output for both sexes and all waves is presented is
Appendix A.6 (page 60). We focus on a few key variables and for ease of exposition refer to
plots of marginal effects of key variables (with solid lines indicating 95% confidence intervals)
in Figures 6-14. All coefficients in the transfer equation are presented as the average marginal
effect on the probability of transfer.11
5.1 Parent leading: Setting a ‘fixed’ transfer level
The coefficients and standard errors for key variables in the parent-leading Stackelberg model are
presented in Tables 4 (boys) and 5 (girls). The estimates for each wave are presented separately.
The first column in each case shows the average marginal effects on the probability of the child
receiving transfers. The second and third columns in each case showing tobit coefficients for
effects on the child’s labour supply assuming first exogeneity (“tobit”) then endogeneity (“FIML
2nd stg”) of labour supply to the parent’s transfer decision. We first discuss results showing
that our model is statistically identified, then show that child labour supply responds to labour
demand conditions in the expected direction, before showing how parents’ responses to these
labour demand conditions means we can reject the parent-leading Stackelberg model.
11Any comparison between sexes is not ‘other things equal’, because of systematic differences in characteristicsbetween these groups. Girls have higher prior educational performance and a lower incidence of special educationalneeds than boys, for example. The results should be interpreted as comparing average marginal effects based ona sample representative of the population of the group and time of interest.
22
Tab
le4:
Estim
ationresults:
Stackelbergmodel
withparentlead
ing(setting‘fixed
’tran
sfer
levelin
advance).
Boy
s
Wave1(age
13-14)
Wave2(age
14-15)
Wave3(age15-16)
Transfer
Work
Hou
rsWork
Hours
Transfer
Work
Hours
WorkHou
rsTransfer
WorkHou
rsWorkHou
rs(F
IML1ststg)
(tobit)
(FIM
L2ndstg)
(FIM
L1st
stg)
(tob
it)
(FIM
L2n
dstg)
(FIM
L1ststg)
(tob
it)
(FIM
L2n
dstg)
Transfer
.-2.890**
*-0.540
.-3.752***
-7.658
**.
-5.074
***
-15.16
8***
.(0.353
)(2.418)
.(0.362)
(3.462
).
(0.408
)(1.869
)
‘Perman
ent’
income
0.11
9**
*.
.0.133***
..
0.06
9*.
.percentile
(0.035)
..
(0.039)
..
(0.040
).
.
IMD
(standardized
)0.008
-0.845*
**
-0.845***
0.014
-0.987***
-0.931
***
0.02
7**
-1.111
***
-0.895
***
(0.011
)(0.209)
(0.211)
(0.010)
(0.262)
(0.271
)(0.010
)(0.303
)(0.327
)
Birth
mon
th-0.000
-0.076
-0.062
0.004*
-0.032
-0.014
0.00
3-0.317
***
-0.287
***
(0.002
)(0.052)
(0.052)
(0.002)
(0.054)
(0.054
)(0.003
)(0.086
)(0.091
)
LEA
age16
-64
0.01
9**
-0.117
-0.167
0.025**
-0.440**
-0.341
*0.02
8***
-0.778
***
-0.502
**unem
ploymentrate
(0.008)
(0.183)
(0.185)
(0.008)
(0.193)
(0.200
)(0.008
)(0.206
)(0.224
)
Priored
ucation
alperform
ance
KS3(age14
)pointscore
..
.-0.003
-0.015
-0.007
-0.001
-0.130
-0.107
(standardized
).
..
(0.007)
(0.232)
(0.235
)(0.008
)(0.268
)(0.280
)
KS2(age11
)pointscore
0.00
30.27
60.261
..
..
..
(standardized
)(0.028
)(0.186
)(0.185)
..
..
..
Parents’ed
ucation
Degree
-0.001
-0.558
-0.626
0.049
-1.642**
-1.416
0.02
6-0.910
-0.594
(0.028
)(0.689)
(0.695)
(0.031)
(0.823)
(0.867
)(0.031
)(0.933
)(0.986
)
A-Levels
-0.008
0.153
0.153
0.028
-0.227
-0.069
0.00
30.04
50.22
4(0.025
)(0.587)
(0.588)
(0.026)
(0.727)
(0.742
)(0.027
)(0.813
)(0.859
)
GCSEs
-0.012
0.106
0.137
0.027
-0.059
0.15
6-0.019
0.74
80.61
5(0.023
)(0.574)
(0.575)
(0.024)
(0.730)
(0.740
)(0.025
)(0.845
)(0.876
)
Jointsign
ificance
oflabou
rmarketvariab
les:χ2 3
14.16
8.64
26.33
24.17
11.28
22.53
38.73
18.47
29.67
(p-value)
(0.002
7)(0.000
0)
(0.0000)
(0.0000)
(0.0000)
(0.000
1)(0.000
0)(0.000
0)(0.000
0)
Observations
4868
4868
4868
Notes:Jointsignifi
cancestatisticsand
p-valu
esare
Wald
test
statistics.
Coeffi
cients
intransfercolu
mn
pre
sente
dasavera
gem
arg
inaleffects
on
thepro
bability
ofa
positivetransferbein
gm
ade(a
spre
sente
din
accom
panyin
ggra
phs).Sta
ndard
errors,clu
stere
dby
school,
inpare
nth
ese
s.Longitudin
alweights
applied.*:p<
0.1;**:p<
0.05;***:p<
0.01.In
dex
ofmultip
ledeprivation
isstandard
ized
by
substra
ctingth
em
ean
and
divid
ingby
standard
deviation.M
onth
-of-birth
within
academ
icyear:
Sept(o
ldest
inyear)
=1,Aug(y
oungest
inyear)
=12.Prioreducationalperform
anceis
standard
ized
avera
gepointsc
ore
atage11
(Key
Sta
ge2)in
wave1,and
age14
(Key
Sta
ge3)in
waves2
and
3.A
dditio
nalcontrols:Pare
nt’sso
cio-econom
icstatu
s,pare
nts’em
ploym
ent,
child’s
prioreducationalperform
ance,re
sidentand
non-resident
siblings,
lone
pare
ntfam
ily,step
fam
ily,pare
nt’sillhealth
lim
itsactivities,
house
hold
own’s
hom
e(m
org
age
oroutright),child’s
specialeducationalneeds(S
EN)classifi
cation,urb
an-rura
lclassifi
cation,child’s
eth
nicity,tim
ing
ofinte
rview.
23
Tab
le5:
Estim
ationresults:
Stackelbergmodel
withparentlead
ing(setting‘fixed
’tran
sfer
levelin
advance):
Girls
Wave1(age
13-14)
Wave2(age
14-15)
Wave3(age15-16)
Transfer
Work
Hou
rsWork
Hours
Transfer
Work
Hours
WorkHou
rsTransfer
WorkHou
rsWorkHou
rs(F
IML1ststg)
(tobit)
(FIM
L2ndstg)
(FIM
L1st
stg)
(tob
it)
(FIM
L2n
dstg)
(FIM
L1ststg)
(tob
it)
(FIM
L2n
dstg)
Transfer
.-0.238**
*7.034**
.-2.796***
-10.64
9***
.-4.549
***
-11.52
1**
.(0.377
)(2.974)
.(0.346)
(1.915
).
(0.393
)(4.514
)
‘Perman
ent’
income
0.082
**
..
0.118***
..
0.10
9**
..
percentile
(0.037)
..
(0.038)
..
(0.047
).
.
IMD
(standardized
)0.01
6*-0.785**
*-0.981***
0.008
-0.337
-0.345
0.02
7**
-0.477
*-0.360
(0.009
)(0.235)
(0.280)
(0.010)
(0.252)
(0.267
)(0.001
1)(0.304
)(0.320
)
Birth
mon
th0.00
2-0.118*
*-0.146**
-0.002
-0.082
-0.103
*0.00
8***
-0.351
***
-0.298
***
(0.002
)(0.063)
(0.070)
(0.002)
(0.051)
(0.054
)(0.003
)(0.080
)(0.084
)
LEA
age16
-64
0.00
5-0.487*
*-0.549**
0.021**
-0.798***
-0.714
***
0.02
7***
-0.902
***
-0.761
***
unem
ploymentrate
(0.007)
(0.203)
(0.234)
(0.009)
(0.212)
(0.232
)(0.008
)(0.220
)(0.252
)
Priored
ucation
alperform
ance
KS3(age14
)pointscore
..
.0.000
0.711
***
0.70
6***
0.00
40.49
9**
0.54
3**
(standardized
).
..
(0.031)
(0.214)
(0.226
)(0.009
)(0.250
)(0.255
)
KS2(age11
)pointscore
-0.006
0.53
4***
0.606***
..
..
..
(standardized
)(0.009
)(0.199
)(0.220)
..
..
..
Parents’ed
ucation
Degree
-0.005
0.26
70.244
0.046
1.744*
2.04
9**
0.07
8**
1.94
8**
2.48
4***
(0.030
)(0.798)
(0.899)
(0.030)
(0.7783)
(0.833
)(0.031
)(0.899
)(0.961
)
A-Levels
-0.43*
-0.290
0.208
-0.003
1.965
***
2.01
1***
0.00
32.29
4***
2.42
1***
(0.025
)(0.694)
(0.803)
(0.026)
(0.723)
(0.757
)(0.025
)(0.791
)(0.816
)
GCSEs
-0.048*
*0.158
0.780
-0.022
2.208
***
2.01
5***
-0.002
2.65
5***
2.61
1***
(0.023
)(0.672)
(0.815)
(0.025)
(0.698)
(0.730
)(0.025
)(0.776
)(0.801
)
Jointsign
ificance
oflabou
rmarketvariab
les:χ2 3
6.81
12.13
29.12
10.48
10.24
22.80
39.57
18.29
24.10
(p-value)
(0.078
2)(0.000
0)
(0.0000)
(0.0149)
(0.0000)
(0.000
0)(0.000
0)(0.000
0)(0.000
0)
Observations
4857
4857
4857
Notes:Jointsignifi
cancestatisticsand
p-valu
esare
Wald
test
statistics.
Coeffi
cients
intransfercolu
mn
pre
sente
dasavera
gem
arg
inaleffects
on
thepro
bability
ofa
positivetransferbein
gm
ade(a
spre
sente
din
accom
panyin
ggra
phs).Sta
ndard
errors,clu
stere
dby
school,
inpare
nth
ese
s.Longitudin
alweights
applied.*:p<
0.1;**:p<
0.05;***:p<
0.01.In
dex
ofmultip
ledeprivation
isstandard
ized
by
substra
ctingth
em
ean
and
divid
ingby
standard
deviation.M
onth
-of-birth
within
academ
icyear:
Sept(o
ldest
inyear)
=1,Aug(y
oungest
inyear)
=12.Prioreducationalperform
anceis
standard
ized
avera
gepointsc
ore
atage11
(Key
Sta
ge2)in
wave1,and
age14
(Key
Sta
ge3)in
waves2
and
3.A
dditio
nalcontrols:Pare
nt’sso
cio-econom
icstatu
s,pare
nts’em
ploym
ent,
child’s
prioreducationalperform
ance,re
sidentand
non-resident
siblings,
lone
pare
ntfam
ily,step
fam
ily,pare
nt’sillhealth
lim
itsactivities,
house
hold
own’s
hom
e(m
org
age
oroutright),child’s
specialeducationalneeds(S
EN)classifi
cation,urb
an-rura
lclassifi
cation,child’s
eth
nicity,tim
ing
ofinte
rview.
24
5.1.1 Identification
The second-stage coefficients representing the child’s labour supply response to parental trans-
fers are identified by the exclusion of parental income from the child’s labour supply equation,
and the probability of positive transfers strictly increasing in parental income. The second line
of coefficients in Tables 4 and 5 (girls)Tables shows that the latter condition clearly holds. A
10 percentile move up the income distribution increases the probability of positive transfer by
around 1 percentage point. This is statistically significant at the 5% level in five out of the six
cases and at the 10% level in the last (boys in wave 3).
5.1.2 Labour demand and child part-time work
Figure 3 shows that being born later in the academic year produces a significant employment
penalty for both boys and girls in wave 3. While the coefficients are negative for earlier waves,
they are not statistically significant. This is consistent with ‘turning 16’ specifically bringing the
significant advantage in accessing the types of jobs systematically vacated by students departing
for university. Figures 4 and 5 show that the IMD and adult unemployment rate always have
a negative effect on predicted working hours. At least one of these coefficients is statistically
significant at the 5% level in all six specifications, and when the wave 1, wave 2 and wave
3 regression models are estimated simultaneously for each gender as ‘Seemingly Unrelated’
systems, the coefficients on both IMD and the unemployment rate are jointly significantly
different from zero for both genders with p < 0.005 in all cases.
5.1.3 Testing the predictions of the Parent-leading Stackelberg model
Inspecting both the child responses to parental transfers, and how parental transfers respond
to labour demand conditions, our results are contrary to the predictions of the parent-leading
Stackelberg model in two ways.
Firstly, Figure 6 shows that boys always reduce their working hours parental transfers increase,
and this effect becomes larger as they become older. Girls also reduce their labour supply in
response to parental transfers at both ages 15 and 16. Indeed, the effect is large enough that, on
average, making transfers would crowd out all the teenager’s working hours. However, labour
supply is increasing in transfer receipt for girls in wave 1. This coefficient has a p-value of
25
Figure 3: Parent-leading Stackelberg
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Effect of being born one month later on predicted
hours of employment
Figure 4: Parent-leading Stackelberg
-1.8
-1.6
-1.4
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Effect of one standard deviation increase in IMD on
predicted hours of employment
Figure 5: Parent-leading Stackelberg
����
�
���
���
������������������������ ��� ������� ������
������������������������������ ����������
������������
����
����
��
����
����
����
����
�� ���� �� ���� �� ����
������ ������ ������
26
0.018 for a two-sided test. Accounting for testing for significant effects on multiple outcomes
for each individual (i.e. their labour supply at each of three points in time), this result remains
significant at the 10% level with a p-value of 0.065.12
It is possible that if girls’ employment in this age-group largely consists of babysitting younger
siblings, payment for this work may be conflated with transfers. We re-estimated the model for
children with no younger siblings (for whom there should be no within-household babysitting
work), and obtained a still positive, but much smaller and statistically insignificant coefficient,
lending limited support to this hypothesis. We present these results and discuss the broader issue
of distinguishing household and market work and present these results in full in Appendix A.5.
The second and most important empirical finding at odds with the predictions of the parent-
leading Stackelberg model is that parents are shown to increase their probability of making cash
transfers in response to the factors which restrict the child’s labour market opportunities. As
we discussed in section 2.4, the parent-leading Stackelberg model would predict that transfers
are held constant or reduced in response to such factors.
Figure 6: Parent-leading Stackelberg
-25
-20
-15
-10
-5
0
5
10
15
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Effect of receiving cash transfer on predicted hours
of employment
Figure 7 shows that parents never significantly reduce their probability of transfer for offspring
who are younger within their cohort, and for girls in the year they turn 16, significantly increase
the probability of transfer. (This result is robust to accounting for multiple testing of the
same hypothesis over each of the three waves, with the initial p-value of 0.0004 adjusted to
12This calculation uses the following Bonferroni Adjustment method, with the adjusted p-value calculated asfollows: padj = 1− (1−p(k))g(k) where g(k) = M1−r(.k). Here M is the number of outcomes being tested (3 eachfor boys and girls), p(k) is the unadjusted p-value for the kth outcome and r(.k) is the mean of the (absolute)pairwise correlations between all the outcomes other than k (i.e. the correlation between girls’ labour supply inwave 2 and wave 3, which is 0.434). See Sankoh, Huque and Duby (1997), pp.2534-2535 for discussion and Akeret al. (2011) for an economics application.
27
Figure 7: Parent-leading Stackelberg
-0.01
-0.005
0
0.005
0.01
0.015
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Average marginal effect of being one month
younger on probability of transfer
Figure 8: Parent-leading Stackelberg
-0.02
-0.01
0
0.01
0.02
0.03
0.04
0.05
0.06
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Average marginal effect of one standard deviation
increase in IMD on probability of transfer
Figure 9: Parent-leading Stackelberg
����
����
����
������������������� ����������� ����������
��������������������������������������
��������������������
�����
�����
�
����
����
����
�� ���� �� ���� �� ����
������ ������ ������
28
0.0018).13 We cannot infer whether the absence of a corresponding effect for boys, for whom
the employment penalty is equally large, is due to systematically different treatment - parents
acting more altruistically, in terms of insuring their consumption, towards their daughters than
sons - or statistical imprecision.
In addition, Figures 8 and 9 show that the local Index of Multiple Deprivation and unemploy-
ment rate always have a positive effect on the probability of transfer. Both are statistically
significant for wave 3, and the unemployment rate coefficient is statistically significant at the
5% level in all other cases except for girls in wave 1. The average marginal effects across the
waves are jointly significantly different from zero for the unemployment rate for both genders
(boys, p = 0.002; girls, p = 0.003), and significant for the IMD for boys (p = 0.035) though
marginally insignificant for girls (p = 0.072).14
These findings show that parents’ behaviour is not time-consistent and not opportunistic in
relation to a ceiling on the child’s hours of employment. We therefore reject a parent-leading
Stackelberg model for the determination of child’s labour supply. This may mean that the
difficulty for the child to obtain and keep a job presents too often and heavy a challenge to
time-consistent behaviour by the parent. Alternatively, the parent may give too low a weight
to his own consumption relative to elements of the child’s utility, and to child’s consumption
relative to human capital, for this decision-making process to make him best off. Either way,
it seems parents prefer to ‘insure’ their child’s consumption against labour market difficulties
outside their control (Ermisch, 2003, p.5).
5.2 Child leading, or parent setting a contingent transfer schedule
We now present results for the child-leading Stackelberg framework, in which shifts in the
child’s labour supply decision identify the parent’s best-response correspondence. Coefficients
and standard errors for key variables are presented in Tables 6 (boys) and 7 (girls), and again
the estimates for each wave are presented separately. The first column in each case shows tobit
coefficients for the effect on the child’s labour supply. The second and third columns in each
case show average marginal effects on the probability of the child receiving transfers assuming
13Using the same Bonferroni Adjustment method as described in footnote 13, with M = 3 and r(.k) = 0.314314These findings show that the IMD is not acting as a proxy for unobserved household level characteristics
with which it is correlated. If this were the case, one would expect a negative sign, because generally the moreaffluent the household, and hence the greater the probability of the transfer, the lower the IMD. As such, we areconfident that by including household level counterparts to the deprivation indicators captured by the IMD, thecoefficients here represent the contextual effect of local labour market conditions.
29
first exogeneity (“probit”) then endogeneity (“FIML 2nd stg”) of transfers to the child’s labour
supply decision. Again, we first present results showing that our model is statistically identified,
before showing how parents’ best response correspondence is a function of household income
and the child’s labour supply.
5.2.1 First stage: Labour supply equation
The parent’s best response correspondence is identified by changes in the child’s labour supply
driven by labour demand factors which we assume have no direct effect on parental transfers,
conditional on the household and area characteristics for which we also control. Figures 10-12,
and the third, fourth and fifth lines of coefficients in Tables 6 and 7 show that the unemployment
rate, IMD and month of birth on labour supply all have the expected negative effect on child
labour supply. Within each gender and wave, they are always jointly significant at the the 5%
level or less. These graphs also support the decision to estimate seperately by wave and sex.
Both boys and girls become progressively more responsive to the adult unemployment rate as
they get older. While the difference in coefficients is not statistically significant, boys appear
better insulated from this wider adult labour market, instead (at least at ages 15 and 16) being
dependent on very localised determinants of opportunities, proxied by the IMD. Turning 16 (in
wave 3) earlier than one’s peers is the only birthday that matters significantly for employment
opportunities. This suggests that jobs this birthday opens up; involving alcohol or catering;
are more common among vacancies created by 18 year-olds leaving for university than those for
which the lifting of the 5-hour Saturday restriction (on one’s 15th birthday) would be employers’
binding concern.
30
Tab
le6:
Estim
ationresults:
Stackelbergmodel
withchildlead
ing,
orparentsettingcontingenttransfer
sched
ule:Boys
Wave1(age
13-14)
Wave2(age
14-15)
Wave3(age15-16)
WorkHours
Transfer
Transfer
WorkHou
rsTransfer
Transfer
Work
Hours
Transfer
Transfer
(FIM
L1ststg)
(probit)
(FIM
L2n
dstg)
(FIM
L1ststg)
(probit)
(FIM
L2ndstg)
(FIM
L1st)
(probit)
(FIM
L2ndstg)
Work
Hou
rs.
-0.019
***
-0.005
.-0.020**
*-0.010*
.-0.019***
-0.007
.(0.002
)(0.006
).
(0.002
)(0.005)
.(0.002)
(0.004)
‘Permanant’
income
0.391
0.10
3**
*0.10
1***
-1.389
0.096*
*0.101**
-0.319
0.059
0.055
percentile
(0.909)
(0.036
)(0.036
)(1.014
)(0.039
)(0.040)
(1.277)
(0.038)
(0.039)
IMD
(standardized
)-0.817*
**.
.-1.017
***
..
-1.158***
..
(0.207
).
.(0.267
).
.(0.317)
..
Birth
month
-0.064
..
-0.037
..
-0.326***
..
(0.052
).
.(0.053
).
.(0.088)
..
LEA
age
16-64
-0.125
..
-0.489
**.
.-0.818***
..
unem
ploymentrate
(0.184
).
.(0.202
).
.(0.222)
..
Prioreducationalperf’
KS3(age14
)point
..
.0.02
3-0.006
-0.006
-0.102
-0.007
-0.006
score
(standardized)
..
.(0.234
)(0.008
)(0.008)
(0.274)
(0.009)
(0.009)
KS2(age11
)point
0.251
0.00
50.00
4.
..
..
.score
(standardized)
(0.186
)(0.008)
(0.008
).
..
..
.
Parents’education
Degree
-0.652
-0.011
-0.016
-1.723
**0.02
30.029
-1.062
0.003
0.006
(0.707
)(0.029)
(0.028
)(0.863
)(0.029
)(0.031)
(0.971)
(0.030)
(0.031)
A-Levels
0.158
-0.014
-0.018
-0.230
0.01
30.014
0.088
-0.008
-0.010
(0.590
)(0.025)
(0.025
)(0.753
)(0.025
)(0.026)
(0.837)
(0.027)
(0.027)
GCSEs
0.165
-0.016
-0.073
-0.018
0.01
70.017
0.870
-0.019
-0.024
(0.572
)(0.024)
(0.024
)(0.745
)(0.023
)(0.024)
(0.864)
(0.025)
(0.004)
Jointsign
ificance
oflabourmarketvariables:
24.49
..
34.85
..
47.94
..
χ2 3(p-value)
(0.000
0)
..
(0.000
0).
.(0.0000)
..
Observation
s48
6848
684868
Notes:Jointsignifi
cancestatisticsand
p-valu
esare
Wald
test
statistics.
Estim
ating
all
six
specifi
cationsjointly
obta
insW
ald
test
statistic
ofχ2 6
=23.64
(p=0.0006)forjointsignifi
canceofchildre
n’s
Work
Hours
on
pare
nts’pro
bability
oftransfer.
Coeffi
cients
intransfercolu
mn
pre
sente
dasavera
ge
marg
inaleffects
on
the
pro
bability
ofa
positive
transferbein
gm
ade
(aspre
sente
din
accom
panyin
ggra
phs).
Sta
ndard
errors,clu
stere
dby
school,
inpare
nth
ese
s.Longitudin
alweights
applied.*:p
<0.1;**:p
<0.05;***:p
<0.01.In
dex
ofmultip
ledeprivation
isstandard
ized
by
substra
cting
them
ean
and
divid
ing
by
standard
deviation.M
onth
-of-birth
within
academ
icyear:
Sept(o
ldest
inyear)
=1,Aug
(youngest
inyear)
=12.Prioreducationalperform
anceis
standard
ized
avera
gepointsc
ore
atage11
(Key
Sta
ge2)in
wave1,and
age14
(Key
Sta
ge3)in
waves2
and
3.A
dditio
nalcontrols:Pare
nt’sso
cio-econom
icstatu
s,pare
nts’em
ploym
ent,
child’s
prioreducationalperform
ance,re
sidentand
non-residentsiblings,
lonepare
ntfam
ily,
step
fam
ily,pare
nt’sillhealth
lim
itsactivities,
house
hold
own’s
hom
e(m
org
ageoroutright),child’s
specialeducationalneeds(S
EN)classifi
cation,urb
an-rura
lclassifi
cation,child’s
eth
nicity,tim
ing
ofinte
rview.
31
Tab
le7:
Estim
ationresults:
Stackelbergmodel
withchildlead
ing,
orparentsettingcontingenttransfer
sched
ule:Girls
Wave1(age13-14)
Wave2(age
14-15)
Wave3(age
15-16)
WorkHou
rsTransfer
Transfer
WorkHou
rsTransfer
Transfer
WorkHou
rsTransfer
Transfer
(FIM
L1ststg)
(probit)
(FIM
L2n
dstg)
(FIM
L1ststg)
(probit)
(FIM
L2n
dstg)
(FIM
L1ststg)
(probit)
(FIM
L2n
dstg)
WorkHou
rs.
-0.016
***
-0.002
.-0.016
***
-0.009
.-0.019
***
-0.015
***
.(0.003
)(0.007
).
(0.002
)(0.006
).
(0.001
)(0.004
)
‘Perman
ant’
income
-0.094
0.087
**0.08
6**
-1.726
*0.09
5**
0.09
8**
-0.572
0.08
7**
0.08
9**
percentile
(0.986
)(0.038
)(0.039
)(1.010
)(0.038
)(0.039
)(1.190
)(0.040
)(0.040
)
IMD
(standardized
)-0.808
***
..
-0.399
..
-0.577
*.
.(0.238)
..
(0.258
).
.(0.320
).
.
Birth
mon
th-0.121
*.
.-0.085
*.
.-0.393
***
..
(0.064)
..
(0.051
).
.(0.083
).
.
LEA
age16
-64
-0.482
**.
.-0.869
***
..
-1.031
***
..
unem
ploymentrate
(0.202
).
.(0.221
).
.(0.233
).
.
Prior
education
alperf’
KS3(age
14)point
..
.0.72
0***
0.00
20.00
20.48
6*0.00
10.00
1score(standardized
).
..
(0.214
)(0.009
)(0.009
)(0.258
)(0.008
)(0.009
)
KS2(age
11)point
0.55
0***
-0.006
-0.007
..
..
..
score(standardized
)(0.201
)(0.009
)(0.009
).
..
..
Parents’ed
ucation
Degree
0.25
3-0.013
-0.014
1.67
5**
0.04
10.04
11.51
00.06
8**
0.07
3**
(0.810)
(0.031
)(0.030
)(0.796
)(0.028
)(0.031
)(0.935
)(0.026
)(0.031
)
A-Levels
-0.190
-0.049
*-0.049
**2.06
5***
-0.002
-0.006
2.30
2***
0.00
70.00
5(0.695)
(0.026
)(0.023
)(0.735
)(0.026
)(0.026
)(0.814
)(0.025
)(0.025
)
GCSEs
0.29
5-0.050
**-0.050
**2.26
8***
-0.018
-0.022
2.65
2***
0.00
2-0.001
(0.676)
(0.024
)(0.023
)(0.706
)(0.025
)(0.025
)(0.798
)(0.002
)(0.025
)
Jointsignificance
oflabou
rmarketvariables:
35.26
..
32.72
..
57.88
..
χ2 3(p-value)
(0.000
0).
.(0.000
0).
.(0.000
0).
.
Observations
4857
4857
4857
Notes:Jointsignifi
cancestatisticsand
p-valu
esare
Wald
test
statistics.
Estim
ating
all
six
specifi
cationsjointly
obta
insW
ald
test
statistic
ofχ2 6
=23.64
(p=0.0006)forjointsignifi
canceofchildre
n’s
Work
Hours
on
pare
nts’pro
bability
oftransfer.
Coeffi
cients
intransfercolu
mn
pre
sente
dasavera
ge
marg
inaleffects
on
the
pro
bability
ofa
positive
transferbein
gm
ade
(aspre
sente
din
accom
panyin
ggra
phs).
Sta
ndard
errors,clu
stere
dby
school,
inpare
nth
ese
s.Longitudin
alweights
applied.*:p
<0.1;**:p
<0.05;***:p
<0.01.In
dex
ofmultip
ledeprivation
isstandard
ized
by
substra
cting
them
ean
and
divid
ing
by
standard
deviation.M
onth
-of-birth
within
academ
icyear:
Sept(o
ldest
inyear)
=1,Aug
(youngest
inyear)
=12.Prioreducationalperform
anceis
standard
ized
avera
gepointsc
ore
atage11
(Key
Sta
ge2)in
wave1,and
age14
(Key
Sta
ge3)in
waves2
and
3.A
dditio
nalcontrols:Pare
nt’sso
cio-econom
icstatu
s,pare
nts’em
ploym
ent,
child’s
prioreducationalperform
ance,re
sidentand
non-residentsiblings,
lonepare
ntfam
ily,
step
fam
ily,pare
nt’sillhealth
lim
itsactivities,
house
hold
own’s
hom
e(m
org
ageoroutright),child’s
specialeducationalneeds(S
EN)classifi
cation,urb
an-rura
lclassifi
cation,child’s
eth
nicity,tim
ing
ofinte
rview.
32
Figure 10: Child-leading Stackelberg
����
�
���
���
������������������������ ��� ������� ������
������������������������������ ����������
������������
����
����
����
��
����
����
����
����
�� ���� �� ���� �� ����
������ ������ ������
Figure 11: Child-leading Stackelberg
-2
-1.5
-1
-0.5
0
0.5
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Effect of one standard deviation increase in IMD on
predicted hours of employment
Figure 12: Child-leading Stackelberg
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0.1
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Effect of being born one month later on predicted
hours of employment
33
5.2.2 Second stage: Parent’s transfer or tax schedule
Figure 13 shows that parent’s transfer probability is always increasing in household income. The
effect is small, akin to a 10 percentage point movement up the income distribution increasing the
probability of transfers by approximately 1 percentage point. We expected a small magnitude for
the income gradient. Earlier work by Dustmann et al. (2009) and Dustmann and Micklewright
(2001) identified a marginal propensity for parents to transfer income to their children of just
0.005 on average, for example. With transfers somewhat inelastic with respect to parental
income, this is consistent with Dustmann et al’s (1996) argument that youths from lower-income
households do not face “disproportionately high financial pressure” to work.15
Figure 14 shows that transfer probability is always decreasing in hours of employment, but this
effect is statistically significant only for girls in wave 3, which is the final year of compulsory
education. This result is robust to adjusting for our testing of multiple null hypotheses (in this
case, that the coefficient on hours of work is zero in each of three time periods). The initial
p-value of 0.0012 becomes 0.0041 after making this adjustment.16 As such, for this final group,
we can firmly reject the null hypothesis that parents do not tax their children. More generally,
when wave 1, wave 2 and wave 3 regression models for each gender are estimated simultaneously
as ‘Seemingly Unrelated’ systems, these ‘tax rate’ coefficients across all three waves are jointly
significantly less than zero at the 1% level for girls (p=0.0038) but and 10% level for boys
(p=0.0563).
These conclusions are clearly affected by the fact that we have information on the extensive
margin only (i.e. whether there is a transfer from parents to children or not). There could be
adjustments on the intensive margin (the amount), but as these changes are not observable our
statistical power to detect parents’ behavioural responses is limited. Our estimates here tell us
simply that the ‘tax rate’ at the intensive margin is sufficiently shallow to mean that increasing
labour supply results in very few cases of parents switching from positive to zero transfers.
Parents are sufficiently altruistic and paternalistic to adopt this decision-making mechanism
in preference to setting a ‘fixed’ transfer level, but since they are not observed to discourage
employment through a heavy ‘tax rate’, these results imply they do not perceive having a job
15A potential alternative strategy was to identify the effect of income changes over time on transfers. Thisproved infeasible due to the differences in income measurement between waves.
16Here we use the same Bonferroni Adjustment method described in footnote 13. The number of outcomesbeing tested is again 3, and r(.k), the mean of the (absolute) pairwise correlations between all the outcomes otherthan k (i.e. the correlation between girls’ cash transfer receipt in wave 1 and wave 2) is 0.452
34
to be particularly damaging to the child’s human capital or educational performance.
Figure 13: Child-leading Stackelberg
-0.05
0
0.05
0.1
0.15
0.2
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Average marginal effect of relative permanent
income on probability of cash transfer
Figure 14: Child-leading Stackelberg
-0.03
-0.025
-0.02
-0.015
-0.01
-0.005
0
0.005
0.01
0.015
Boys Girls Boys Girls Boys Girls
Wave 1 Wave 2 Wave 3
Ave' marginal effect one hour p.week increase in
child employment on prob' of cash transfer
Nevertheless, we might expect the productivity of study time, in terms of educational perfor-
mance measured at the end of school year 11, to be highest in that final year of compulsory
schooling; and the amount of study time undertaken also to be highest close to these final ex-
ams, resulting in the greatest scope for study time being crowded out. Certainly, it has been
shown that the net effect of in-school employment is more detrimental to academic performance
in closer proximity to examinations (Ruhm, 1997; Payne, 2004). In this case, we would expect
parents’ incentive to use their financial resources to deter in-school employment, and hence our
ability to detect such a tax schedule, to be highest in the final year of compulsory education.
The results in Figure 14 are consistent with this prediction, though also imply that, at least for
work undertaken during this final year, parents perceive the net effect of in-school labour supply
35
to be more detrimental for girls than boys.17 This may be because girls dominate employment
in ‘child-oriented’ roles such as babysitting (Kooreman, 2009), which may contribute more to
family-oriented and less to career-oriented skills and preferences, resulting in less favourable
long-run labour market outcomes due to changes in fertility decisions, for example (Erdogan
et al., 2012). An alternative explanation is that parents have a higher relative preference for
education-oriented human capital in their daughters than sons, perhaps reflecting women’s
higher wage returns to educational qualifications (Machin, 2005; Bonesrønning, 2010).18
6 Conclusion
Employment experience for teenagers in compulsory education has a potentially significant
impact on educational performance and subsequent academic and labour market opportunities.
This paper has focused on the idea that parents can use financial incentives to alter their child’s
decision to work, and developed a theoretical model for the interaction determining the financial
support provided by parents and part-time employment taken by children. We motivate two
alternative Stackelberg frameworks. In the first (parent-leading), parents set their transfer level
in advance, anticipating the child’s best-response. In the second (child leading) parents set a
contingent transfer schedule for each level of labour supply, effectively taxing their children’s
earnings.
We test our model using data on a cohort of English teenagers between 2004 and 2006, estimat-
ing cross-sectional models for the same sample of individuals at ages 14, 15, and 16. Holding
constant the institutional arrangements, education system and time-invariant unobserved char-
acteristics, we identify significant differences in the determinants of selection into employment
17The results in Holford (2016), estimated using the same cohort of individuals, suggest that such an expectationwould be rational. Most previous literature either focuses exclusively on males, in order that longer-run labourmarket outcomes may be considered without fertility considerations, or does not allow for heterogeneous effectsby gender. The extant evidence is inconclusive. Ruhm (1997) suggests the human capital benefit of in-schoolemployment is greater for females but D’Amico (1984) shows the opposite conclusion, while Rothstein (2007)shows similar and small negative effects for both genders.
18In this approach, we have implicitly assumed the same decision-making process is adopted by all households.However, as we note in section 2.3, the less paternalistic and more selfish is the parent, the more likely it is thatshe will prefer to set a fixed transfer level. Heterogeneous preferences may therefore lead to heterogeneity in thedecision-making mechanism. Recognizing this, strictly speaking, our results show only that a sufficiently largeproportion of parents adopt the child-leading model that the parent-leading model can be rejected in statisticalterms. In Appendix A.1 (page 44), we derive the predictions of the model assuming the parent gains no utilityfrom human capital, and show estimates for two subpopulations of potential policy interest: households in whichthe parent (i) would like the child to leave full-time education at the minimum age and (ii) has no qualifications.Though hampered by small samples, these estimates provide tentative evidence for these households moving fromthe parent-leading to child-leading mechanism between ages 14 and 16.
36
as the children age, but do not detect significant differences in the propensity of parents to ‘tax’
their children’s earnings over this time.
Estimates from the empirical counterpart to the parent-leading framework show that parents
use their relative financial power to ‘insure’ their child’s consumption against factors beyond
their control. Rather than holding transfers constant in response to factors affecting the child’s
labour market opportunities, or reducing them in response to a ceiling being placed on their
hours of employment, parents raise transfers instead. This suggests that parents would like
their children to provide for their own consumption, but will not penalize them if this proves
difficult. We also demonstrate a very shallow income gradient in parent’s willingness to make
cash transfers to the child. These results suggest that parents of children who otherwise would
leave education at 16 can and will accommodate the two additional years of schooling which,
from 2013-2014, have been becoming compulsory in the UK, as an ‘extension to childhood’,
rather than impose a financial burden on an ‘independent adult’. This should ensure the
benefits to extended compulsory education will not be hindered by financial pressures to take
employment.
These results are inconsistent with a model in which parents set a ‘fixed’ transfer level. We
therefore reject the parent-leading Stackelberg framework. The findings are in line with a
situation in which parents set a contingent transfer schedule for each possible level of the child’s
labour supply. In the language of game theory this can be interpreted as the parent - whose
financial resources afford the choice between the two decision-making mechanisms - being better
off by ceding the Stackelberg first-move advantage to the child. The child’s human capital is a
public good, valued by both the parent and the child but potentially eroded by long hours of
employment. In this framework, by taxing the child’s earnings the parent forces the child to
internalise this social cost. This cannot be achieved with a fixed lump-sum transfer.
We find that the probability of receiving transfers is always decreasing in the child’s hours of
employment, so parents do tax their children, though this is only significant for girls in the
final year of compulsory schooling. Further work should try to explore whether this gender
distinction is due to gender differences in parental perceptions of the education production
function or gender-specific parental preferences for education-oriented human capital.
Our results suggest there is little case for more restrictive regulation of in-school employment
on equity grounds. There is strong positive selection into employment by prior educational
37
performance, and the shallow income gradient in parental transfers reveals no evidence that
consumption pressures drive low socio-economic status children into employment. Nevertheless,
from the perspective of a social planner, it may still be the case that parents underestimate the
human capital cost of employment, leading to sub-optimal outcomes.
38
References
Aker, J., Boumnijel, R., McClelland, A., Tierney, N., 2011. Zap it to me: The impacts of a
mobile cash transfer program. Center for Global Development Working Paper 268.
Antman, F., 2012. Elderly care and intrafamily resource allocation when children migrate.
Journal of Human Resources 47 (2), 331–363.
Apps, P., Rees, R., 1996. Labour supply, household production and intra-family welfare distri-
bution. Journal of Public Economics 60, 199–219.
Apps, P., Rees, R., 2009. Public Economics and the Household. Cambridge University Press,
Cambridge, United Kingdom.
Ashworth, A., Ulph, D., 2011. Household models. In: Brown, C. (Ed.), Taxation and Labour
Supply. Vol. 33. Allen and Unwin, pp. 117–133.
Becker, G., 1974. A theory of social interactions. Journal of Political Economy 82 (6), 1063–1093.
Becker, G., 1981. Altruism in the family and selfishness in the market place. Economica 48 (189),
1–15.
Bell, C., Fesbach, H., Schneider, M., 2012. Raising juveniles. Games and Economic Behavior
74, 32–51.
Bergstrom, T. C., 1989. A fresh look at the rotten kid theorem - and other household mysteries.
Journal of Political Economy 97 (5), 1138–1159.
Bingley, P., Walker, I., 2013. There’s no such thing as a free lunch: Evidence on altruism
and agency from household expenditure responses to child nutrition programs. Review of
Economics of the Household 5, 371–392.
Blundell, R., Chiappori, P.-A., Meghir, C., 2005. Collective labor supply with children. Journal
of Political Economy 113 (6), 1277–1306.
Bonesrønning, H., 2010. Are parental effort allocations biased by gender? Education Economics
18 (3), 253–268.
Bourguignon, F. J., 1999. The cost of children: may the collective approach to household
behaviour help? Journal of Population Economics 12 (4), 503–522.
39
Browning, M., Chiappori, P., 1998. Efficient intra-household allocations: A general characteri-
sation. Econometrica 66 (6), 1241–1278.
Bruce, N., Waldman, M., 1991. Transfers in kind: why they can be efficient and nonpaternalistic.
American Economic Review 81 (5), 1345–1351.
Bucciol, A., Veronesi, M., 2014. Teaching children to save: What is the best strategy for lifetime
savings? Journal of Economic Psychology 45, 1–17.
Bureau of Labor Statistics, 2000. Report on the youth labor force.
Burton, P., Phipps, S., Curtis, L., 2002. All in the family: A simultaneous model of parenting
style and child conduct. American Economic Review 92 (2), 368–372.
Buscha, F., Maurel, A., Page, L., Speckesser, S., 2011. The effect of high school employment on
educational attainment: A conditional difference-in-differences approach. Oxford Bulletin of
Economics and Statistics 74 (3), 380–396.
Collins, W., 1990. Parent-child relationships in the transition to adolescence: Continuity and
change in interaction, affect, and cognition. In: Montemayor, R., Adams, G., Gullotta, T.
(Eds.), From childhood to adolescence: A transitional period? Advances in adolescent devel-
opment. Vol. 2. Sage, Thousand Oaks, CA, pp. 85–106.
Collins, W., Gleason, T., Sesma, Jr, A., 1997. Internalization, autonomy, and relationships: De-
velopment during adolescence. In: Grusec, J., Kuczynski, L. (Eds.), Parenting and childrens
internalization of values: A handbook of contemporary theory. Wiley, New York, pp. 78–99.
Collins, W., Luebker, C., 1994. Parent and adolescent expectancies: Individual and relationship
significance. In: Smetana, J. (Ed.), Beliefs about parenting: Origins and developmental im-
plications. New Directions for Child Development. Vol. 66. Jossey-Bass, San Francisco, CA,
pp. 65–80.
Conti, G., Pudney, S. E., 2011. Survey design and the analysis of satisfaction. Review of Eco-
nomics and Statistics 93 (3), 1087–1093.
Crawford, C., Dearden, L., Greaves, E., 2013. When you are born matters: Evidence for Eng-
land. Institute for Fiscal Studies Report R80.
40
D’Amico, R., 1984. Does employment during high school impair academic progress? Sociology
of Education 57, 152–164.
Dauphin, A., El Lahga, A.-R., Fortin, B., Lacroix, G., 2011. Are children decision-makers within
the household? Economic Journal 121, 871–903.
Department for Children, Schools and Families, 2009. Guidance on the employment of chil-
dren. Department for Children, Schools and Families: https://www.education.gov.uk/
publications/eOrderingDownload/Child_employment09.pdf, accessed 24-08-2012.
Department for Education, National Centre for Social Research, May 2013. Longitudinal Study
of Young People in England: Waves One to Seven, 2004-2010: Secure Access. computer file,
2nd Edition. Colchester, Essex: UK Data Archive [distributor], May 2013. SN: 7104.
Dustmann, C., Micklewright, J., 2001. Intra-household transfers and the part-time work of
children. CEPR discussion papers 2796.
Dustmann, C., Micklewright, J., Rajah, N., Smith, S., 1996. Earning and learning: Educational
policy and the growrth of part-time work by full-time pupils. Fiscal Studies 17 (1), 79–103.
Dustmann, C., Micklewright, J., van Soest, A., 2009. In-school labour supply, parental transfers
and wages. Empirical Economics 37 (1), 201–218.
Erdogan, Z., Jacobson, J., Kooreman, P., 2012. Do babysitters have more kids? The effects of
teenage work experiences on adult outcomes. IZA Discussion Papers 6856.
Ermisch, J., 2003. An Economic Analysis of the Family. Princeton University Press, Princeton
NJ.
Gong, T., 2009. Do parental transfers reduce youths’ incentives to work? Labour: Review of
Labour Economics and Industrial Relations 23 (4), 653–676.
Holford, A., 2016. Youth employment and academic performance: production functions and
policy effects. IZA Discussion Papers 10009.
Hotz, V. J., Xu, L. C., Tienda, M., Ahituv, A., 2002. Are there returns to the wages of young
men from working while in school? The Review of Economics and Statistics 84 (2), 221–236.
Howieson, C., McKechnie, J., Semple, S., 2006. The nature and implications of the part-time
employment of secondary school pupils. Scottish Executive Social Research.
41
James, N., Silva, P., Smith, P., 2010. Implementing a strategy to compensate for missing data
in the Longitudinal Study of Young People in England. Tech. rep., ORC International for the
Department for Children, Schools, and Families.
Kalenkoski, C. M., Pabilonia, S. W., 2010. Parental transfers, student achievement and the
labor supply of college students. Journal of Population Economics 23 (2), 469–496.
Konrad, K. A., Lommerud, K. E., 1995. Family policy with non-cooperative families. Scandi-
navian Journal of Economics 97 (4), 581–601.
Kooreman, P., 2009. The early inception of labor market gender differences. Labour Economics
16, 135–139.
Kooreman, P., Kapteyn, A., 1990. On the empirical implementation of some game theoretic
models of household labour supply. Journal of Human Resources 25, 584–598.
Leonard, R., 104. Reading Cournot, reading Nash: The creation and stabilisation of the Nash
equilibrium. Economic Journal 104, 492–511.
Leuthold, J., 1968. An empirical study of formula income transfers and the work decision of the
poor. Journal of Human Resources 3, 312–323.
Light, A., 2001. In-school work experience and the returns to schooling. Journal of Labor Eco-
nomics 19 (1), 65–93.
Lillydahl, J. H., 1990. Academic achievement and part-time employment of high school students.
Journal of Economic Education 21 (3), 307–316.
Lindbeck, A., Weibull, J., 1988. Altruism and time consistency: the economics of fait accompli.
Journal of Political Economy 96 (6), 1165–1182.
Lundberg, S., Romich, J. L., Tsang, K. P., 2009. Decision-making by children. Review of Eco-
nomics of the Household 7, 1–30.
Machin, S., 2005. Wage inequality since 1975. In: Dickens, R., Gregg, P., Wadsworth, J. (Eds.),
The Labour Market under New Labour: The state of working Britain. Palgrave Macmillan,
pp. 191–200.
McElroy, M., Horney, M., 1981. Nash bargained household decisions: towards a generalization
of the theory of demand. International Economic Review 22 (2), 333–349.
42
Montmarquette, C., Viennot-Briot, N., Dagenais, M., 2007. Dropout, school peformance and
working while in school. Review of Economics and Statistics 89 (4), 752–760.
Oettinger, G. S., 1999. Does high school employment affect high school academic performance.
Industrial and Labor Relations Review 53 (1), 136–151.
Office of the Deputy Prime Minister, 2004. The English Indices of Deprivation 2004: Sum-
mary (revised). http://webarchive.nationalarchives.gov.uk/20120919132719/www.
communities.gov.uk/documents/communities/pdf/131206.pdf, accessed 25-07-2013.
Payne, J., 2004. The impact of part-time jobs in years 12 and 13 on qualification achievement.
British Educational Research Journal 29 (4), 599–611.
Plug, E., 2001. Season of birth, schooling and earnings. Journal of Economic Psychology 22,
641–660.
Rothstein, D., 2007. High school employment and youths’ acadmic achievement. Journal of
Human Resources 42 (1), 194–213.
Ruhm, C. J., 1997. Is high school employment consumption or investment? Journal of Labor
Economics 15 (4), 735–776.
Sankoh, A., Huque, M., Dubey, S., 1997. Some comments on frequently used multiple endpoint
adjustment methods in clinical trials. Statistics in Medicine 16, 2529–2542.
Schmidt, K., 1993. Reputation and equilibrium characterization in repeated games with con-
flicting interests. Econometrica 61 (2), 325–351.
Shim, S., Barber, B., Card, N., Xiao, A., Serido, J., 2010. Financial socialization of first year
college students. the roles of parents, work, and education. Journal of Youth and Adolescence
39, 1457–1470.
Steinberg, L., 2001. We know some things: Adolescent-parent relationships in retrospect and
prospect. Journal of Research on Adolescence 11, 1–19.
Tyler, J. H., 2003. Using state child labor laws to identify the effect of school-year work on high
school achievement. Journal of Labor Economics 21 (2), 381–408.
Wolff, F.-C., 2006. Parental transfers and the labor supply of children. Journal of Population
Economics 19 (4), 853–877.
43
A Appendix
A.1 Parents with no preference for human capital
The theoretical predictions in the main body of this paper are driven by the assumption that
the parent is paternalistic; she cares more about the child’s human capital than does the child
himself. This enables us to reject the Nash equilibrium on welfare grounds, and our empirical
results suggest that parents’ relative preference for human capital is sufficiently high for them to
prefer to set a contingent transfer schedule based on the child’s labour market earnings, rather
than a fixed transfer level. Thus, it is the parent’s preference for human capital that drives her
to ‘insure’ the child’s consumption against labour market factors outside the child’s control.
In this appendix, we show how the predictions of the model change when the parent does not
care about the child’s human capital. We show that in this case, in common with conventional
non-cooperative games, she will prefer to move first and set a fixed transfer level, thus no
longer insuring the child’s consumption. We test the same predictions of the parent-leading
Stackelberg model for two subpopulations that may be expected to have a lower preference
for human capital than the pooled population: parents who declare that they would like their
child to leave full-time education at age 16; and parents with no educational qualifications. We
find tentative evidence that parents in these groups transition from the parent-leading to the
child-leading model between the child’s ages of 14 and 16, suggesting either that the parent’s
relative preference for human capital, or their perception of the negative human capital effect
of in-school employment, increases over time.
A.1.1 Parent’s maximisation problem
The parent’s maximisation problem under a Nash equilibrium or when she acts as a Stackelberg
follower, in either case taking the child’s action as given, is now as follows:
maxt
Up = Up(P (M − t), C(wL+ t), κp(X)) (A1)
Because in these cases, she takes the child’s action (L) as given, her optimality condition is
unchanged:
UpC = U
pP (A2)
44
Acting as a Stackelberg leader, the parent must account for the indirect effects of t on the utility
from consumption, but no longer from the child’s human capital, caused by the child’s labour
supply response to t:
maxt
Up = Up(P (M − t), C(wL∗(t) + t), κp(X)) (A3)
The parent’s optimality condition as a Stackelberg leader is, hence:
UpC(1 +
∂L∗
∂t· w) = U
pP (A4)
Since the child’s maximisation problems and optimality conditions are unchanged, he will still
reduce labour supply in response to increasing transfers, so ∂L∗
∂t< 0. This means, at the Nash
equilibrium level of transfers it holds that:
UpC(1 +
∂L∗
∂t· w) < U
pP (A5)
From this position, a parent who does not care about the child’s human capital will reduce
transfers, thereby increasing her own consumption (and reducing its marginal utility UpP ) and
reducing the child’s (increasing his marginal utility U cC). This means that the parent-leading
Stackelberg equilibrium will occur at lower transfers and higher labour supply than the Nash
equilibrium.
This situation can again be represented graphically, as shown in Figure 15. There are two key
differences from the situation portrayed in Figure 1. Firstly, since the parent now will now never
subsidise employment in order to encourage human capital accumulation, there is no range of
labour supply over which the parent increases transfers in response to increased labour supply.
This means that ∂t∗
∂L< 0 for all L, not only in equilibrium, so the parent’s BR is downward
sloping over its entire range. Secondly, the parent’s bliss point is now above the child’s best-
response correspondence, meaning that the parent-leading Stackelberg model is strictly preferred
to the Nash equilibrium, in turn strictly preferred to the child-leading Stackelberg equilibrium.
The child’s preference ordering is precisely the opposite (Since the child’s optimality condition
is unchanged, the child-leading Stackelberg equilibrium will still occur at higher labour supply
and lower transfers than the Nash equilibrium).
45
Figure 15: Agents’ optimisation problem and the three equilibria when the parent has nopreference for child’s human capital
Parent bliss point
Human
cap’ max’
0 → t t
L
↑
0
L
= t*(L, M, X)
= L*(t, w, X)
Child bliss
point at
t=∞
Nash equilibrium
Child-leading
Stackelberg
equilibrium
Parent-leading
Stackelberg
equilibrium
Increasing child’s utility
Increasing parent’s utility
A.2 Estimates
Our estimates of the parent-leading model are shown in Tables A1 and A2, and the child-leading
model in Tables A3 and A4. In each case, the results for the group whose parents would like
them to leave full-time education at 16 is shown in the upper panel, those whose parents have
no qualifications in the lower panel, and in both cases boys first.
In the parent-leading model we must first note that in three cases (for sons of parents who would
like them to leave at 16, in waves 2 and 3, and daughters of parents with no qualifications, in
wave 1), the coefficient on ‘permanent’ income is negative. Though in these cases the effect
is not significant, the positive coefficients obtained in the majority of other cases also fail to
obtain significance. This means that the effect of transfers on labour supply in the second stage
of these models is not identified. Nevertheless, by focusing on the first stage results we are able
to test the prediction of the parent leading model that the parent reduces or does not alter
transfers in response to labour market difficulties faced by the child.
For the group whose parents would like them to leave full-time education, among boys we fail to
reject this hypothesis in wave 1. Indeed, there is strong evidence that parents are less likely to
46
give transfers to children who are younger within the school year (having controlled for month
of interview, including a dummy for interview after the child’s birthday). In wave 2, we again
fail to reject the hypothesis, but in wave 3, it is firmly rejected due the statistically significant
positive coefficient on the LAD unemployment rate.
For girls, in wave 1 we find a positive coefficient on IMD, which is significant at the 5% level, but
conversely, a negative coefficient on the cusp of significance at the 10% level, for the unemploy-
ment rate, though it should also be noted that for this sample, the effect of this unemployment
rate on hours of employment is positive but not statistically different from zero. This suggests
tentative grounds to reject the parent-leading model, but in wave 2 there is no such evidence. In
wave 3, the coefficients on IMD, birth month and unemployment are all positive, with the coef-
ficient on birth month individually and the three coefficients jointly significant at the 10% level,
meaning that there are again grounds to reject the parent-leading model. (Note the especially
small sample for this specification, with which a 10% rejection is relatively more acceptable).
For those whose parents have no qualifications the picture is simpler. For boys the parent-
leading model cannot be rejected in wave 1, there are grounds for rejection in wave 2 (a 5%
significant positive coefficient on unemployment, though the three variables together are not
collectively significant), and a firmer rejection in wave 3, due to the 1% significant coefficient
on IMD. For girls, we always reject the parent-leading model, firmly in waves 1 and 3, less so
in wave 2, where the coefficient on birth month is in fact negative (with a t-statistic of 1), but
on the other two is positive and on the cusp of significance).
Moving to the child-leading model, the counterpart testable prediction is that we should obtain
a negative coefficient on Work Hours in the parent’s transfer equation. The failure to obtain
statistical significance in the main body of the paper with the larger sample suggests obtaining
robust inference on this hypothesis is challenging.
Nevertheless, at face value, we observe a positive (though never signifcant at any level) coefficient
on Work Hours in the Transfer equation in three cases, all for girls; in waves 1 and 2 among those
whose parents wanted them to leave education, and in wave 1 among those whose parents have
no qualifications. In the former two cases, we failed to reject the parent-leading model, while
in the last the parent leading model was rejected somewhat strongly. We observe a negative
coefficient at the 5% level only for girls whose parents want them to leave, in wave 3 of the
survey, which is consistent with the results in the main body of this paper.
47
Altogether, therefore, the results presented in this section show that, even if parents’ stated
attitude places a low weight on educational human capital (since they would prefer their child
to leave full-time education as soon as possible), by the time the child reaches the final year of
compulsory schooling their observed behaviour is sufficient to reject the parent-leading model
and consistent with setting a contingent transfer level, or ‘taxing’ their children. This means
that despite their stated preference, they still care sufficiently about the effects of labour supply
on immediate educational performance to prefer this strategy over setting a ‘fixed’ transfer
level. However, there is some tentative evidence that, unlike the general population, while the
children are aged 14 or 15, parents may prefer to set a fixed transfer level.
Our interpretation of this finding is that, at age 14, this subpopulation cares less about human
capital than the general population, but this preference increases over time as the importance
of educational qualifications for early labour market opportunities becomes more salient.
48
Table A1: Estimation results for subpopulation of parents who would like child to leave full-timeeducation at age 16: Parent-leading model
Boys
Wave 1 (age 14) Wave 2 (age 15) Wave 3 (age 16)
Transfer Work Hours Transfer Work Hours Transfer Work Hours(1st stg) (2nd stg) (1st stg) (2nd stg) (1st stg) (2nd stg)
Transfer . -6.375 . -10.249 . -15.917. (4.259) . (10.034) . (5.685)
‘Permanent’ income 0.110 . -0.228 . -0.121 .percentile (0.076) . (0.082) . (0.076) .
IMD (standardized) -0.004 -0.613 0.021 -0.961* 0.011 0.186(0.019) (0.416) (0.021) (0.536) (0.022) (0.659)
Birth month -0.017*** -0.101 0.005 -0.047 -0.009 -0.587***(0.005) (0.134) (0.005) (0.139) (0.006) (0.218)
LAD age 16-64 0.022 0.141 0.016 -0.246 0.056*** -0.968unemployment rate (0.017) (0.367) (0.016) (0.465) (0.019) (0.619)
Joint significance of labour 13.20 2.74 4.55 4.90 15.94 9.61market variables: χ2
3(p-value) (0.004) (0.433) (0.208) (0.179) (0.001) (0.022)
Observations 981 1077 1034
Girls
Wave 1 (age 14) Wave 2 (age 15) Wave 3 (age 16)
Transfer Work Hours Transfer Work Hours Transfer Work Hours(1st stg) (2nd stg) (1st stg) (2nd stg) (1st stg) (2nd stg)
Transfer . 2.297 . -19.712*** . -10.103. (9.038) . (3.204) . (9.606)
‘Permanent’ income 0.103 . 0.020 . -0.151 .percentile (0.139) . (0.083) . (0.133) .
IMD (standardized) 0.057** -0.368 0.021 0.240 0.023 -1.618*(0.024) (0.889) (0.029) (1.004) (0.036) (0.963)
Birth month 0.006 -0.256 -0.002 -0.126 0.018* -0.287(0.008) (0.205) (0.006) (0.204) (0.009) (0.302)
LAD age 16-64 -0.0451 0.297 0.037 -0.525 0.039 -0.357unemployment rate (0.027) (0.804) (0.022) (0.864) (0.027) (0.736)
Joint significance of labour 4.94 2.00 5.05 0.72 7.44 4.58market variables: χ2
3(p-value) (0.176) (0.573) (0.168) (0.868) (0.059) (0.205)
Observations 485 482 447
Notes: 1: p-value = 0.101. So on margin of statistical significance. Joint significance statistics and p-values are Wald test statistics. Coefficientsin transfer column presented as average marginal effects on the probability of a positive transfer being made. Standard errors, clustered byschool, in parentheses. Longitudinal weights applied. *: p ≤ 0.1; **: p ≤ 0.05; ***: p ≤ 0.01. Index of multiple deprivation is standardizedby substracting the mean and dividing by standard deviation. Month-of-birth within academic year: Sept (oldest in year) = 1, Aug (youngestin year) = 12. Prior educational performance is standardized average point score at age 11 (Key Stage 2) in wave 1, and age 14 (Key Stage3) in waves 2 and 3. Additional controls: Parent’s highest qualification, parent’s socio-economic status, parents’ employment, child’s prioreducational performance, resident and non-resident siblings, lone parent family, child’s special educational needs (SEN) classification, urban-ruralclassification, child’s ethnicity, timing of interview.
49
Table A2: Estimation results for subpopulation of parents who have no qualifications: Parent-leading model
Boys
Wave 1 (age 14) Wave 2 (age 15) Wave 3 (age 16)
Transfer Work Hours Transfer Work Hours Transfer Work Hours(1st stg) (2nd stg) (1st stg) (2nd stg) (1st stg) (2nd stg)
Transfer . 11.663 . 12.086 . -13.033. (5.097) . (17.988) . (7.684)
‘Permanent’ income 0.128 . 0.046 . 0.073 .percentile (0.081) . (0.144) . (0.120) .
IMD (standardized) 0.016 -2.358*** 0.004 -0.378 0.065*** -0.427(0.023) (0.690) (0.026) (1.267) (0.026) (1.219)
Birth month -0.002 -0.041 0.001 -0.164 -0.004 -0.622(0.006) (0.200) (0.007) (0.357) (0.008) (0.431)
LAD age 16-64 0.010 0.130 0.048** -1.950 0.034 0.079unemployment rate (0.018) (0.547) (0.024) (2.047) (0.022) (0.801)
Joint significance of labour 1.85 12.34 5.47 1.45 16.26 3.12market variables: χ2
3(p-value) (0.605) (0.006) (0.141) (0.694) (0.001) (0.374)
Observations 806 727 731
Girls
Wave 1 (age 14) Wave 2 (age 15) Wave 3 (age 16)
Transfer Work Hours Transfer Work Hours Transfer Work Hours(1st stg) (2nd stg) (1st stg) (2nd stg) (1st stg) (2nd stg)
Transfer . -8.078 . -17.382 . 2.324. (10.599) . (11.736) . (5.555)
‘Permanent’ income -0.090 . 0.206 . 0.291** .percentile (0.110) . (0.122) . (0.126) .
IMD (standardized) 0.046** -0.892 0.044 0.105 0.093*** -1.293(0.019) (0.736) (0.025) (1.290) (0.025) (1.195)
Birth month 0.001 -0.480** -0.007 -0.053 0.015* -0.417(0.006) (0.209) (0.007) (0.364) (0.009) (0.349)
LAD age 16-64 0.016 0.503 0.034 -0.263 0.016 -1.104unemployment rate (0.016) (0.621) (0.022) (1.095) (0.021) (0.834)
Joint significance of labour 10.47 7.13 11.71 0.009 27.42 6.19market variables: χ2
3(p-value) (0.015) (0.068) (0.008) (0.993) (0.000) (0.103)
Observations 891 801 797
Notes: 1: p-value = 0.101. So on margin of statistical significance. Joint significance statistics and p-values are Wald test statistics. Coefficientsin transfer column presented as average marginal effects on the probability of a positive transfer being made. Standard errors, clustered byschool, in parentheses. Longitudinal weights applied. *: p ≤ 0.1; **: p ≤ 0.05; ***: p ≤ 0.01. Index of multiple deprivation is standardizedby substracting the mean and dividing by standard deviation. Month-of-birth within academic year: Sept (oldest in year) = 1, Aug (youngestin year) = 12. Prior educational performance is standardized average point score at age 11 (Key Stage 2) in wave 1, and age 14 (Key Stage3) in waves 2 and 3. Additional controls: Parent’s highest qualification, parent’s socio-economic status, parents’ employment, child’s prioreducational performance, resident and non-resident siblings, lone parent family, child’s special educational needs (SEN) classification, urban-ruralclassification, child’s ethnicity, timing of interview.
50
Table A3: Estimation results for subpopulation of parents who would like child to leave full-timeeducation at age 16:: Child-leading model
Boys
Wave 1 (age 14) Wave 2 (age 15) Wave 3 (age 16)
Transfer Work Hours Transfer Work Hours Transfer Work Hours(1st stg) (2nd stg) (1st stg) (2nd stg) (1st stg) (2nd stg)
Work Hours . -0.012 . -0.011 . -0.002. (0.013) . (0.009) . (0.010)
‘Permanant’ income -0.985 0.097 0.080 -0.040 4.150* -0.097percentile (1.746) (0.077) (1.927) (0.079) (2.477) (-0.002)
IMD (standardized) -0.592 . -1.099** . 0.187 .(0.407) . (0.507) . (0.651) .
Birth month -0.008 . -0.088 . -0.476** .(0.137) . (0.124) . (0.210) .
LAD age 16-64 0.027 . -0.405 . -1.499** .unemployment rate (0.368) . (0.415) . (0.652) .
Joint significance of labour 2.57 . 10.12 . 10.18 .market variables: χ2
3(p-value) (0.463) . (0.018) . (0.017) .
Observations 981 1077 1034
Girls
Wave 1 (age 14) Wave 2 (age 15) Wave 3 (age 16)
Transfer Work Hours Transfer Work Hours Transfer Work Hours(1st stg) (2nd stg) (1st stg) (2nd stg) (1st stg) (2nd stg)
Work Hours . 0.018 . 0.006 . -0.029***. (0.021) . (0.167) . (0.009)
‘Permanant’ income 4.624 0.044 1.616 0.125 6.094* -0.046percentile (2.522) (0.021) (2.833) (0.104) (3.123) (0.113)
IMD (standardized) 0.217 . -0.278 . -1.596* .(0.734) . (0.942) . (0.906) .
Birth month -0.331* . -0.116 . -0.569** .(0.184) . (0.195) . (0.259) .
LAD age 16-64 -0.085 . -0.586 . -0.875 .unemployment rate (0.732) . (0.853) . (0.707) .
Joint significance of labour 3.28 . 1.09 . 13.86 .market variables: χ2
3(p-value) (0.350) . (0.780) . (0.003) .
Observations 485 482 447
Notes: Joint significance statistics and p-values are Wald test statistics. Coefficients in transfer column presented as average marginal effectson the probability of a positive transfer being made. Standard errors, clustered by school, in parentheses. Longitudinal weights applied. *:p ≤ 0.1; **: p ≤ 0.05; ***: p ≤ 0.01. Index of multiple deprivation is standardized by substracting the mean and dividing by standard deviation.Month-of-birth within academic year: Sept (oldest in year) = 1, Aug (youngest in year) = 12. Prior educational performance is standardizedaverage point score at age 11 (Key Stage 2) in wave 1, and age 14 (Key Stage 3) in waves 2 and 3. Additional controls: Parent’s highestqualification, Parent’s socio-economic status, parents’ employment, child’s prior educational performance, resident and non-resident siblings,lone parent family, child’s special educational needs (SEN) classification, urban-rural classification, child’s ethnicity, timing of interview.
51
Table A4: Estimation results for subpopulation of parents who have no qualifications: Child-leading model
Subpopulation: Parent has no qualifications
Boys
Wave 1 (age 14) Wave 2 (age 15) Wave 3 (age 16)
Transfer Work Hours Transfer Work Hours Transfer Work Hours(1st stg) (2nd stg) (1st stg) (2nd stg) (1st stg) (2nd stg)
Work Hours . -0.021 . -0.021* . -0.017. (0.173) . (0.013) . (0.013)
‘Permanant’ income 5.523** 0.062 4.234 0.053 2.526 0.062percentile (2.763) (0.100) (5.181) (0.109) (5.347) (0.114)
IMD (standardized) -1.901*** . -0.116 . -1.048 .(0.526) . (1.114) . (1.187) .
Birth month -0.082 . -0.139 . -0.556 .(0.145) . (0.314) . (0.417) .
LAD age 16-64 0.366 . -0.950 . 0.432 .unemployment rate (0.534) . (1.013) . (0.826) .
Joint significance of labour 14.08 . 1.69 . 4.85 .market variables: χ2
3(p-value) (0.003) . (0.639) . (0.183) .
Observations 806 727 731
Girls
Wave 1 (age 14) Wave 2 (age 15) Wave 3 (age 16)
Transfer Work Hours Transfer Work Hours Transfer Work Hours(1st stg) (2nd stg) (1st stg) (2nd stg) (1st stg) (2nd stg)
Work Hours . 0.009 . -0.002 . -0.012. (0.021) . (0.020) . (0.021)
‘Permanant’ income 5.665* -0.090 -2.885 0.219 -0.583 0.225*percentile (2.830) (0.095) (5.134) (0.110) (5.208) (0.130)
IMD (standardized) -0.984 . -0.417 . -1.118 .(0.711) . (1.233) . (1.446) .
Birth month -0.486** . 0.033 . -0.398 .(0.197) . (0.332) . (0.357) .
LAD age 16-64 0.451 . -0.614 . -0.992 .unemployment rate (0.571) . (1.175) . (0.842) .
Joint significance of labour 8.35 . 0.63 . 4.40 .market variables: χ2
3(p-value) (0.039) . (0.8898) . (0.2211) .
Observations 891 801 797
Notes: Joint significance statistics and p-values are Wald test statistics. Coefficients in transfer column presented as average marginal effectson the probability of a positive transfer being made. Standard errors, clustered by school, in parentheses. Longitudinal weights applied. *:p ≤ 0.1; **: p ≤ 0.05; ***: p ≤ 0.01. Index of multiple deprivation is standardized by substracting the mean and dividing by standard deviation.Month-of-birth within academic year: Sept (oldest in year) = 1, Aug (youngest in year) = 12. Prior educational performance is standardizedaverage point score at age 11 (Key Stage 2) in wave 1, and age 14 (Key Stage 3) in waves 2 and 3. Additional controls: Parent’s highestqualification, Parent’s socio-economic status, parents’ employment, child’s prior educational performance, resident and non-resident siblings,lone parent family, child’s special educational needs (SEN) classification, urban-rural classification, child’s ethnicity, timing of interview.
52
A.3 Observed social interactions
This section derives the child’s labour supply response to unearned income in the Nash or
parent-leading Stackelberg models and the parent’s transfer response to child labour supply in
the Nash or child-leading Stackelberg model. These parameters are derived from the implicit
functions which are the first-order conditions in equations (4) and (9).
A.3.1 The labour supply response to unearned income
The child’s first-order condition is:
[.] = w · U cC +HL · U c
H = 0 (A6)
Using implicit differentiation, the parameter of interest is equal to:
∂L
∂t= −
[∂[.]∂t
]
[∂[.]∂L
](A7)
= −∂[wUc
C ]
∂t+
∂[UcHHL]
∂t∂[wUc
C]
∂L+
∂[UcH
HL]
∂L
The elements of this expression are as follows:
• .∂[w·Uc
C ]∂t
= wU cCC : w is constant, and increasing t by 1 increases consumption by 1.
• .∂[Uc
HHL]∂t
= 0: Human capital production function is invariant to transfers, and marginal
utility from human capital and consumption are separable.
• .∂[wUc
C ]∂L
= w2U cCC : w is constant, and increasing L by 1 changes marginal utility of
consumption as though consumption increased by w.
• .∂[Uc
HHL]∂L
= U cHH(HL)
2 + U cH ·HLL: Using the product rule, bearing in mind that U c
H is
itself a function of H.
Multiplying and dividing the whole expression by w, the parameter of interest is:
∂L
∂t= −
1
w·
w2 · U cCC
w2 · U cCC + (HL)2 · U c
HH + U cHHLL
(A8)
53
This means that − 1w
< ∂L∂t
< 0. The child always withdraws labour earnings at a slower rate
than transfers are received. Hence, in the Nash or parent-leading Stackelberg model, child’s
consumption is strictly increasing in cash transfers.
A.3.2 The ‘tax’ rate
The parent’s first-order condition is:
[.] = −UpP + U
pC = 0 (A9)
The parameter of interest is therefore equal to:
∂t
∂L= −
[∂[.]∂L
]
[∂[.]∂t
](A10)
= −−
∂UpP
∂L+
∂UpC
∂L
−∂U
pP
∂t+
∂UpC
∂t
The elements of this expression are as follows:
• .∂U
pP
∂L= 0: Child’s labour supply has no effect on parent’s consumption.
• .∂U
pC
∂L= w · U
pCC : Increasing consumption by 1 means U
pC changes by U
pCC , so at the
margin, increasing consumption by w means UpC changes by w.U
pCC .
• .∂U
pP
∂t= −U
pPP : Increasing t by 1 has same effect as reducing own consumption by 1.
• .∂U
pC
∂t= U
pCC : Increasing t by 1 has same effect as increasing child’s consumption by 1.
This means the parameter of interest is:
∂t
∂L= −w ·
UpCC
UpPP + U
pCC
(A11)
This can be expressed as the ‘tax’ rate, in response to earnings rather than labour supply:
∂t
∂L·1
w= −
UpCC
UpPP + U
pCC
(A12)
The denominator here is more negative than numerator, so the expression is strictly between
zero and minus 1: the ‘tax rate’ is strictly between 0 and 100%. This means that in the Nash
54
or child-leading Stackelberg model, the parent withdraws transfers at a rate slower than wages
are earned, so the child’s consumption is strictly increasing in his labour supply.
55
A.4 Differentiated labour markets by gender
A.4.1 Job types
The LSYPE does not provide information on the types of jobs teenagers do. Howieson et al.
(2006) indicate the extent to which boys’ and girls’ labour markets are differentiated in Scot-
land, where the education system, labour market conditions and legislation are very similar to
England. The largest categories of employment among those aged 14 and under are newspaper
delivery (32% of workers, among which boys outnumber girls by a factor of five) and babysitting
(11% of workers, among which girls outnumber boys by a factor of six). In both cases, the gen-
der bias is partly driven by differences in physical and mental attributes affecting individuals’
ability to do the job well, but also by social norms for gender roles. This suggests that the size
and male dominance of the market for newspaper delivery is responsible for the higher male
employment at age 13-14, when few other jobs are available.
The largest categories of employment among 15 and 16 year olds are “cafes and restaurants”
(17.5%), “other shops” (14%) and “chain stores” (12%), with females more heavily concentrated
in these industries than males. Jobs in these categories are also likely to be restricted to 16 year-
olds in law (due to health and safety requirements in food preparation or the sale of alcohol) or
15 year-olds in practice (because this enables an eight hour shift to be supplied on a Saturday).
A.4.2 Gender roles and ethnicity
Table A5 shows that the overall trends in employment are driven by the majority white ethnic
group, who also have the highest employment levels for both sexes at every wave. There is
considerable variation in employment rates and trajectories among the other ethnic groups.
This will reflect a combination of socio-cultural norms, prior educational performance, poorer
labour market opportunities and active labour market discrimination, but also rigid gender
roles within certain ethnic groups which make employment among girls exceptionally rare. In
particular, Pakistani and Bangladeshi girls are almost completely excluded from the labour
market. This makes it sensible to estimate our models separately for boys and girls.19
19In both the parent-leading and child-leading specifications several ethnicity dummies are statistically signif-icant in both transfer and labour supply equations. Ethnicity therefore makes a difference in levels. However,by re-estimating on white students only and obtaining very similar results, we find no evidence for significantethnicity interactions with the parameters of interest.
56
Table A5: Employment rates by wave, sex and ethnicity
Wave 1 Wave 2 Wave 3
Boys Girls Boys Girls Boys Girls
White 25.7% 21.3% *** 31.4% 30.3% 30.8% 34.8% ###
Mixed 15.7 % 16.0% 15.6% 19.9% 24.5% 20.5%
Indian 9.4% 4.4% *** 12.1% 5.0% *** 10.1% 3.4% ***
Pakistani 7.0% <1.5% *** 6.4% <1.5% *** 8.0% <1.5% ***
Bangladeshi 4.5% <1.5% *** 5.8% <1.5% *** 7.2% 3.1% *
Black Caribbean 11.2% 9.6% 11.5% 10.7% 12.1% 14.7%
Black African 7.8% 4.7% 10.1% 5.2% 6.0% 8.7%
Other 15.3% 9.8% 16.7% 8.1% 12.8% 11.1%
Symbols: *, **, ***: Boys’ employment higher than girls’ at 5%, 1% and 0.1% levels respectively. #, ##, ###: Girls’ employmenthigher than boys’ at 5%, 1% and 0.1% levels respectively. Standard errors clustered by school. Population proportions calculated using finalprobability weights.
57
A.5 Distinguishing household and market work
We do not observe job types but have assumed that the question on labour supply is interpreted
as referring to market work. We assume that this includes informal employment in the parent’s
own business, and that survey respondents distinguish the payments they receive for this work
from pocket money or allowances. However, it is possible that teenagers may regard babysitting
within as well as outside the household as ‘paid work’, yet double-count the payment they receive
as part of their reported transfers. If this is true the observed labour supply response (parent-
leading model) or parent’s ‘tax rate’ (child-leading model) will be positively biased (less negative
or more positive than the true effect).
We test this hypothesis by re-estimating the model for children with no younger siblings, for
whom there is no possibility of babysitting within the household. The key results are shown
in table A6. The coefficient on receiving cash transfers for girls in wave 1 of the parent-
leading model becomes less positive and insignificantly different from zero. The coefficient for
boys has become more negative (though the change is less drastic) and is now statistically
significant. These results lend qualified support to our initial rejection of the parent-leading
Stackelberg model based on 13-14 year-old girls’ apparent positive labour supply response to
parental transfers being due to this babysitting hypothesis. However, the parent’s significant
positive transfer response to both the IMD and month of birth means these results are still
inconsistent with the parent-leading Stackelberg model.
In the child-leading model, the wave 1 labour supply of daughters with no younger siblings
appears to be associated with a small ‘subsidy’, or increase in probability of transfers. This
finding at face value rejects the babysitting hypothesis, but this coefficient is very imprecisely
estimated and not statistically different from zero or the pooled-sample coefficient. The coeffi-
cients on cash transfers or labour supply do not change significantly either for boys or for older
girls (for whom we assume this matters less - see discussion in Appendix A.4, above), though
the conclusion drawn would change: statistically significant ‘tax rates’ are now observed in wave
2 rather than wave 3. It is possible that as children with ‘no younger siblings’ are more likely to
have ‘some older siblings’ than the pooled sample, this reflects parents having a softer attitude
towards labour supply in the final year of school, once they have seen how older siblings coped.
58
Table A6: Estimation results for sample members with no younger siblings.
Boys
Parent-leading Child-leading
Wave 1 Wave 1 Wave 2 Wave 3
Transfer Work Hours Work Hours Transfer Work Hours Transfer Work Hours Transfer(FIML 1st) (FIML 2nd) (FIML 1st) (FIML 2nd) (FIML 1st) (FIML 2nd) (FIML 1st) (FIML 2nd)
Transfer . -3.072* . . . . . .. (1.787) . . . . . .
Work Hours . . . -0.009 . -0.013* . -0.003. . . (0.008) . (0.007) . (0.007)
‘Permanent’ income 0.121** . -1.244 0.095* -1.231 0.118** 3.412 -0.081percentile (0.059) . (1.494) (0.056) (1.760) (0.059) (2.072) (0.065)
IMD (standardized) 0.009 -0.942*** -0.994*** . -0.688* . -0.936* .(0.016) (0.367) (0.342) . (0.394) . (0.539) .
Birth month -0.004 -0.113 -0.104 . -0.104* . -0.233* .(0.004) (0.085) (0.088) . (0.096) . (0.147) .
LEA age 16-64 0.016 0.028 -0.004 . -0.583* . -0.678* .unemployment rate (0.012) (0.271) (0.275) . (0.312) . (0.373) .
Joint significance of labourmarket variables: χ2
3. 11.88 13.03 . 11.59 . 11.98 .
(p-value) . (0.008) (0.005) . (0.009) . (0.008) .
Observations 1881 1881 1881 1881
Girls
Parent-leading Child-leading
Wave 1 Wave 1 Wave 2 Wave 3
Transfer Work Hours Work Hours Transfer Work Hours Transfer Work Hours Transfer(FIML 1st) (FIML 2nd) (FIML 1st) (FIML 2nd) (FIML 1st) (FIML 2nd) (FIML 1st) (FIML 2nd)
Transfer . 2.592 . . . . . .. (3.714) . . . . . .
Work Hours . . . 0.010 . -0.018* . -0.009. . . (0.010) . (0.009) . (0.007)
‘Permanent’ income 0.088* . 1.004 0.066 -1.478 0.117** -1.168 0.083percentile (0.052) . (1.435) (0.054) (1.502) (0.009) (1.680) (0.057)
IMD (standardized) 0.028** -0.907** -0.697* . -0.240 . -1.107* .(0.014) (0.444) (0.395) . (0.395) . (0.490) .
Birth month 0.007* -0.279*** -0.223*** . -0.098 . -0.386** .(0.004) (0.093) (0.086) . (0.080) . (0.123) .
LEA age 16-64 -0.006 -0.558* -0.596** . -1.074*** . -1.092*** .unemployment rate (0.010) (0.312) (0.295) . (0.329) . (0.317) .
Joint significance of labourmarket variables: χ2
3. 23.17 22.38 . 16.35 . 28.09 .
(p-value) . (0.000) (0.000) . (0.001) . (0.000) .
Observations 1915 1915 1915 1916
Notes: Joint significance statistics and p-values are Wald test statistics. Coefficients in transfer column presented as average marginal effectson the probability of a positive transfer being made presented in accompanying graphs). Standard errors, clustered by school, in parentheses.Longitudinal weights applied. *: p < 0.1; **: p < 0.05; ***: p < 0.01. Index of multiple deprivation is standardized by substractingthe mean and dividing by standard deviation. Month-of-birth within academic year: Sept (oldest in year) = 1, Aug (youngest in year) =12. Prior educational performance is Additional controls: Parent’s socio-economic status, parents’ employment, child’s prior educationalperformance (standardized average point score at age 11 - Key Stage 2 - in wave 1 and age 14 - Key Stage 3 - in waves 2 and 3), residentand non-resident siblings, lone parent family, step family, parent’s ill health limits activities, household own’s home (morgage or outright),child’s special educational needs (SEN) classification, urban-rural classification, child’s ethnicity, timing of interview.
59
A.6 Complete estimation output
The complete maximum likelihood estimation output for the parent-leading Stackelberg model
is shown in Tables A7-A8, and the child-leading Stackelberg model in Tables A9-A10, starting
overleaf. Tables 4-7 in the main body presented average marginal effects on the probability of
transfer for a set of key variables. Here we show the probit linear latent regression coefficients for
both dependent and all explanatory variables. We omit the output for the baseline specifications
assuming that transfers and labour supply are exogenous in the second stage of the parent-
leading and child-leading specifications respectively.
60
Tab
leA7:
Com
plete
max
imum
likelihoodestimationresults:
Parent-lead
ingmodel
Boys
Girls
Wave1
Wave2
Wave3
Wave1
Wave2
Wave3
tran
sfer
WorkHou
rstran
sfer
Work
Hou
rstran
sfer
WorkHou
rstran
sfer
Work
Hours
transfer
Work
Hours
transfer
Work
Hours
Transfer
-0.852
-7.299
**-15.14
6***
6.950**
-10.233***
-11.126**
(2.213
)(3.676
)(1.877
)(2.984)
(2.043)
(5.293)
‘Perman
ent’
income
0.44
6***
0.45
4***
0.22
8*0.31
4**
0.424***
0.387**
percentile
(0.132
)(0.134
)(0.133
)(0.144
)(0.137)
(0.169)
IMD
(standardized
)0.03
0-0.854
***
0.04
9-0.952
***
0.08
8**
-0.917
***
0.06
1*-0.952***
0.029
-0.330
0.094**
-0.356
(0.041
)(0.211
)(0.035
)(0.271
)(0.035
)(0.327
)(0.036
)(0.279)
(0.037)
(0.265)
(0.039)
(0.322)
Birth
mon
th-0.002
-0.075
0.01
5*-0.016
0.01
1-0.290
***
0.00
9-0.146**
-0.007
-0.097*
0.029***
-0.300***
(0.009
)(0.052
)(0.008
)(0.054
)(0.010
)(0.091
)(0.009
)(0.070)
(0.008)
(0.054)
(0.010)
(0.085)
LEA
age16
-64
0.07
3**
-0.155
0.08
7***
-0.357
*0.09
5***
-0.523
**0.02
1-0.552**
0.075**
-0.690***
0.096***
-0.742***
unem
ploymentrate
(0.029
)(0.185
)(0.028
)(0.201
)(0.026
)(0.225
)(0.029
)(0.234)
(0.031)
(0.229)
(0.029)
(0.263)
Prior
education
alperform
ance
KS3(age
14)pointscore
-0.011
-0.021
-0.004
-0.131
0.000
0.720***
0.014
0.529**
(standardized
)(0.027
)(0.236
)(0.029
)(0.281
)(0.030)
(0.225)
(0.032)
(0.254)
KS2(age
11)pointscore
0.01
00.26
6-0.025
0.604***
(standardized
)(0.031
)(0.186
)(0.033
)(0.218)
Parents’em
ployment:
TwoFT
-0.028
1.10
1-0.226
**1.85
0**
0.07
81.76
1-0.106
1.807**
-0.070
1.794**
0.056
3.500**
(0.126
)(0.785
)(0.112
)(0.849
)(0.178
)(1.584
)(0.120
)(0.835)
(0.112)
(0.809)
(0.191)
(1.653)
oneFT,on
ePT
-0.019
1.23
4-0.102
2.35
0***
0.08
61.04
2-0.060
1.227
-0.069
1.653**
0.114
3.135*
(0.121
)(0.762
)(0.108
)(0.781
)(0.178
)(1.604
)(0.116
)(0.868)
(0.105)
(0.781)
(0.186)
(1.657)
OneFT
only
-0.012
0.74
5-0.101
1.35
6*0.06
30.70
60.01
40.811
-0.081
0.326
0.078
1.352
(0.104
)(0.684
)(0.099
)(0.755
)(0.158
)(1.426
)(0.099
)(0.772)
(0.097)
(0.764)
(0.172)
(1.521)
One/tw
oPT,noFT
0.14
71.12
40.03
70.96
50.21
01.57
40.04
50.541
-0.055
0.530
0.095
0.602
(0.117
)(0.760
)(0.109
)(0.835
)(0.157
)(1.464
)(0.107
)(0.872)
(0.104)
(0.774)
(0.172)
(1.485)
Non
-resident
-0.015
-0.012
-0.004
0.20
1-0.020
0.22
7-0.017
0.318
-0.024
0.438**
-0.044
-0.297
siblings
(0.031
)(0.166
)(0.022
)(0.162
)(0.023
)(0.217
)(0.033
)(0.238)
(0.023)
(0.183)
(0.027)
(0.218)
Residentsiblings
-0.055
**0.55
5***
-0.031
0.32
5**
-0.064
***
0.42
4**
-0.115
***
1.038***
-0.098***
0.581***
-0.075***
0.318
(0.024
)(0.138
)(0.023
)(0.159
)(0.022
)(0.202
)(0.025
)(0.227)
(0.028)
(0.201)
(0.024)
(0.211)
Parents’ed
ucation
Degree
-0.002
-0.583
0.16
9-1.444
*0.08
8-0.559
-0.021
0.279
0.167
2.127**
0.276**
2.504**
(0.107
)(0.695
)(0.106
)(0.871
)(0.102
)(0.985
)(0.115
)(0.896)
(0.109)
(0.830)
(0.110)
(0.975)
A-Levels
-0.030
0.15
80.09
6-0.114
0.01
10.18
8-0.167
*0.214
-0.014
1.979***
0.012
2.376***
(0.092
)(0.589
)(0.089
)(0.745
)(0.088
)(0.859
)(0.096
)(0.799)
(0.093)
(0.752)
(0.090)
(0.812)
GCSEs
-0.046
0.13
40.09
30.14
9-0.063
0.63
0-0.186
**0.761
-0.082
2.052***
-0.006
2.632***
(0.090
)(0.575
)(0.082
)(0.741
)(0.082
)(0.876
)(0.089
)(0.809)
(0.088)
(0.725)
(0.089)
(0.798)
Other
qualification
s-0.364
*1.75
2-0.123
2.13
3-0.305
0.01
7-0.177
0.061
-0.140
-0.447
0.136
2.264
(0.212
)(1.356
)(0.193
)(1.450
)(0.188
)(2.149
)(0.203
)(1.728)
(0.185)
(1.638)
(0.222)
(2.007)
Parents’SES
NSSEC
1-2
0.03
0-0.675
*0.00
8-0.210
0.07
60.37
40.14
2**
-1.105**
-0.007
-0.487
0.052
0.161
(0.070
)(0.385
)(0.062
)(0.376
)(0.060
)(0.533
)(0.064
)(0.529)
(0.061)
(0.422)
(0.063)
(0.487)
NSSEC
30.13
7-0.706
0.03
4-1.578
**0.19
5**
-0.412
0.03
7-0.753
0.024
-0.800
0.039
0.831
(0.091
)(0.578
)(0.091
)(0.650
)(0.096
)(0.859
)(0.092
)(0.661)
(0.095)
(0.709)
(0.097)
(0.762)
NSSEC
4(self-
-0.235
***
0.04
9-0.128
-0.114
-0.001
0.47
40.06
90.740
0.036
1.385**
0.021
1.610**
employed)
(0.085
)(0.464
)(0.085
)(0.599
)(0.095
)(0.785
)(0.080
)(0.577)
(0.089)
(0.639)
(0.092)
(0.712)
NSSEC8imp
-0.068
-0.530
0.06
8-1.336
-0.047
-0.287
-0.003
-0.408
0.028
0.033
0.155
0.946
(0.136
)(0.963
)(0.139
)(1.177
)(0.140
)(1.294
)(0.129
)(1.381)
(0.128)
(1.254)
(0.158)
(1.410)
Lon
eparent
0.05
4-0.061
0.08
10.30
1-0.014
-0.535
-0.027
1.144*
0.116
0.977*
0.152*
1.726***
(0.081
)(0.493
)(0.072
)(0.524
)(0.074
)(0.670
)(0.086
)(0.632)
(0.074)
(0.524)
(0.078)
(0.628)
Observations
4854
4854
4854
4846
4846
4846
Notes:*:p<
0.1;**:p<
0.05
***:p<
0.01.Sta
ndard
errors,clu
stere
dby
school,
inpare
nth
ese
s.Longitudin
alweights
applied.
61
Tab
leA8:
Com
plete
max
imum
likelihoodestimationresults:
Parent-lead
ingmodel
(continued
)
Boys
Girls
Wave1
Wave2
Wave3
Wave1
Wave2
Wave3
tran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rs
You
th’s
ethnicity
Mixed
race
-0.026
-3.013***
0.100
-2.401**
0.032
-2.048**
-0.112
-0.136
-0.053
-0.149
-0.117
-1.567
(0.132)
(0.940)
(0.121)
(1.148)
(0.119)
(0.940)
(0.128)
(0.922)
(0.142)
(0.889)
(0.132)
(1.093)
Indian
-0.549***
-3.725***
-0.423***
-5.036***
-0.355***
-7.292***
-0.633***
-3.738**
-0.576***
-8.796***
-0.313***
-10.703***
(0.096)
(0.836)
(0.093)
(1.062)
(0.101)
(1.293)
(0.111)
(1.472)
(0.102)
(1.100)
(0.106)
(1.395)
Pak
istanior
0.143
-6.182***
0.113
-6.802***
0.297**
-6.688***
0.072
-9.158***
0.083
-12.876***
0.061
-11.102***
Ban
glad
eshi
(0.112)
(0.880)
(0.119)
(1.029)
(0.124)
(1.370)
(0.113)
(1.487)
(0.109)
(2.097)
(0.116)
(1.478)
Black
African
or-0.068
-2.049*
0.043
-2.456*
-0.035
-4.948**
-0.031
-1.726*
-0.078
-3.010***
0.242**
-3.252***
Caribbean
(0.127)
(1.240)
(0.142)
(1.325)
(0.136)
(1.944)
(0.108)
(0.940)
(0.111)
(1.039)
(0.120)
(1.078)
Other
0.005
0.163
-0.171
-2.334
0.060
-3.420*
-0.344*
-1.340
-0.518***
-4.484**
-0.505***
-7.416***
(0.206)
(1.203)
(0.190)
(1.498)
(0.201)
(1.988)
(0.182)
(1.566)
(0.158)
(1.925)
(0.159)
(2.148)
SpecialEd’Needs
-0.097
-0.432
-0.026
-0.914*
-0.047
-1.071*
-0.064
-0.335
0.003
-0.965
-0.054
-2.141***
(0.066)
(0.438)
(0.064)
(0.517)
(0.067)
(0.638)
(0.084)
(0.687)
(0.080)
(0.642)
(0.081)
(0.698)
Tow
nor
Village
-0.007
-0.154
-0.089
0.865*
-0.012
1.264**
-0.117*
1.580***
-0.217***
1.479***
-0.193***
1.663***
(0.071)
(0.377)
(0.066)
(0.463)
(0.064)
(0.537)
(0.069)
(0.521)
(0.066)
(0.420)
(0.064)
(0.533)
Isolated
area
-0.256*
-0.110
-0.224*
0.554
-0.454***
0.392
-0.045
1.130
-0.219*
2.749***
-0.047
3.054***
(0.145)
(0.743)
(0.120)
(0.965)
(0.124)
(1.115)
(0.127)
(0.952)
(0.128)
(0.732)
(0.141)
(0.884)
Illhealthlimits
0.154**
0.147
0.083
0.101
0.058
-0.009
-0.049
-0.189
0.039
-0.819**
0.026
-0.073
parent’sactivities
(0.068)
(0.405)
(0.068)
(0.503)
(0.062)
(0.552)
(0.062)
(0.483)
(0.059)
(0.413)
(0.065)
(0.513)
Hom
eow
ner
(mortgage
-0.063
0.058
0.105
0.513
-0.092
-0.150
-0.031
-0.580
-0.051
0.044
0.058
0.329
oroutright)
(0.066)
(0.430)
(0.066)
(0.517)
(0.065)
(0.622)
(0.074)
(0.519)
(0.068)
(0.485)
(0.073)
(0.548)
StepFam
ily
0.106
-0.099
0.081
-0.079
-0.007
-0.050
0.129
-0.094
-0.003
0.018
-0.084
-0.317
(0.075)
(0.472)
(0.072)
(0.523)
(0.067)
(0.645)
(0.081)
(0.607)
(0.075)
(0.506)
(0.072)
(0.592)
Not
Greater
Lon
don
-0.161*
2.081***
-0.064
1.234
-0.099
2.073**
0.078
1.908***
0.105
2.689***
0.116
1.239*
(0.086)
(0.681)
(0.096)
(0.767)
(0.093)
(0.831)
(0.087)
(0.721)
(0.078)
(0.778)
(0.084)
(0.733)
Tim
eof
interview
March
-0.093
0.175
-0.042
0.225
-0.037
-0.480
-0.032
0.502
0.043
1.835**
-0.073
-0.732*
(0.065)
(0.376)
(0.127)
(0.868)
(0.054)
(0.452)
(0.068)
(0.512)
(0.125)
(0.873)
(0.054)
(0.406)
April
-0.056
0.353
-0.110
0.248
-0.055
0.712
-0.029
0.391
0.013
2.202**
0.002
1.093
(0.067)
(0.433)
(0.126)
(0.890)
(0.081)
(0.727)
(0.073)
(0.532)
(0.124)
(0.869)
(0.077)
(0.692)
May
-0.057
0.908*
-0.123
0.159
-0.287**
-2.224
-0.068
1.305**
0.068
1.895**
-0.311***
-1.883
(0.088)
(0.519)
(0.129)
(0.934)
(0.142)
(1.497)
(0.092)
(0.641)
(0.126)
(0.898)
(0.118)
(1.180)
June
-0.089
1.115
-0.170
1.185
-0.241
-1.256
-0.061
1.142
-0.074
1.440
-0.288
2.429
(0.111)
(0.873)
(0.137)
(0.989)
(0.206)
(2.172)
(0.127)
(1.096)
(0.133)
(0.982)
(0.208)
(1.692)
July
0.194
0.654
0.061
1.370
-0.091
.0.029
0.380
-0.041
2.140
0.092
1.595
(0.152)
(0.855)
(0.204)
(1.318)
(0.637)
.(0.158)
(1.120)
(0.205)
(1.683)
(0.625)
(2.683)
After
birthday
-0.166**
0.182
0.071
0.386
0.029
-0.308
0.052
-0.123
-0.063
-0.460
0.087
-0.410
(0.073)
(0.486)
(0.079)
(0.589)
(0.075)
(0.709)
(0.079)
(0.596)
(0.079)
(0.497)
(0.077)
(0.635)
Observations
4854
485
44854
4846
4846
4846
Notes:*:p<
0.1;**:p<
0.05
***:p<
0.01.Sta
ndard
errors,clu
stere
dby
school,
inpare
nth
ese
s.Longitudin
alweights
applied.
62
Tab
leA9:
Com
plete
max
imum
likelihoodestimationresults:
Child-leadingmodel
Boys
Girls
Wave1
Wave2
Wave3
Wave1
Wave2
Wave3
tran
sfer
WorkHours
transfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rs
Work
Hou
rs-0.018
-0.033*
-0.023
-0.008
-0.031
-0.055***
(0.022)
(0.017)
(0.015)
(0.026)
(0.022)
(0.017)
‘Perman
ent’
income
0.381***
0.341
0.347**
-1.299
0.181
-0.375
0.333**
-0.124
0.355**
-1.664*
0.322**
-0.641
percentile
(0.135)
(0.915)
(0.137)
(1.016)
(0.130)
(1.285)
(0.148)
(0.988)
(0.140)
(1.007)
(0.146)
(1.194)
IMD
(standardized)
-0.830***
-1.034***
-1.178***
-0.786***
-0.389
-0.582*
(0.207)
(0.268)
(0.317)
(0.238)
(0.257)
(0.320)
Birth
mon
th-0.076
-0.040
-0.331***
-0.120*
-0.078
-0.393***
(0.052)
(0.053)
(0.089)
(0.064
)(0.051)
(0.083)
LEA
age16-64
-0.122
-0.499**
-0.843***
-0.489**
-0.849***
-1.018***
unem
ploymentrate
(0.185)
(0.202)
(0.223)
(0.203)
(0.219)
(0.232)
Prioreducation
alperform
ance
KS3(age
14)pointscore
-0.021
0.014
-0.019
-0.125
0.007
0.744***
0.002
0.483*
(standardized)
(0.028)
(0.234)
(0.030)
(0.275)
(0.032)
(0.213)
(0.033)
(0.258)
KS2(age
11)pointscore
0.015
0.256
-0.026
0.553***
(standardized)
(0.032)
(0.187)
(0.033)
(0.200)
Parents’em
ployment:
TwoFT
-0.017
0.984
-0.166
2.430***
0.036
1.365
-0.122
1.814**
-0.023
2.118**
0.087
3.371**
(0.126)
(0.839)
(0.110)
(0.907)
(0.178)
(1.508)
(0.119)
(0.801)
(0.113
)(0.896)
(0.193)
(1.617)
OneFT,on
ePT
-0.019
1.147
-0.064
2.649***
0.014
0.610
-0.080
1.302
-0.044
1.958**
0.121
2.855*
(0.122)
(0.792)
(0.104)
(0.837)
(0.180)
(1.494)
(0.115)
(0.809)
(0.104
)(0.832)
(0.188)
(1.617)
OneFT
only
-0.005
0.679
-0.067
1.632**
0.033
0.344
0.002
0.980
-0.071
0.632
0.041
1.211
(0.105)
(0.707)
(0.098)
(0.783)
(0.159)
(1.343)
(0.096)
(0.685)
(0.099
)(0.778)
(0.172)
(1.477)
One/tw
oPT,noFT
0.141
1.078
0.050
0.891
0.168
0.668
0.040
0.711
-0.052
0.657
0.021
0.482
(0.119)
(0.763)
(0.107)
(0.834)
(0.157)
(1.391)
(0.108)
(0.750)
(0.105
)(0.761)
(0.176)
(1.452)
Non
-resident
-0.015
-0.004
0.003
0.207
-0.019
0.342
-0.017
0.252
-0.023
0.538***
-0.038
-0.159
siblings
(0.030)
(0.169)
(0.022)
(0.166)
(0.023)
(0.212)
(0.034)
(0.202)
(0.023)
(0.184)
(0.026)
(0.227)
Residentsiblings
-0.052**
0.566***
-0.020
0.410**
-0.055**
0.723***
-0.111***
0.750***
-0.097***
0.902***
-0.065***
0.587***
(0.024)
(0.135)
(0.023)
(0.160)
(0.023)
(0.198)
(0.025)
(0.157)
(0.028
)(0.175)
(0.025)
(0.168)
Parents’education
Degree
-0.037
-0.608
0.098
-1.742**
0.021
-1.024
-0.056
0.300
0.149
1.762**
0.264**
1.594*
(0.107)
(0.709)
(0.105)
(0.865)
(0.103)
(0.970)
(0.116)
(0.809)
(0.111
)(0.795)
(0.112)
(0.932)
A-Levels
-0.061
0.164
0.047
-0.275
-0.032
0.055
-0.187*
-0.167
-0.020
2.033***
0.018
2.309***
(0.093)
(0.593)
(0.088)
(0.755)
(0.089)
(0.836)
(0.097)
(0.695)
(0.094
)(0.733)
(0.092)
(0.813)
GCSEs
-0.068
0.165
0.060
-0.021
-0.079
0.867
-0.194**
0.295
-0.079
2.289***
-0.002
2.688***
(0.091)
(0.573)
(0.082)
(0.746)
(0.082)
(0.863)
(0.089)
(0.676)
(0.090
)(0.704)
(0.090)
(0.796)
Other
qualification
s-0.371*
1.917
-0.130
2.452*
-0.292
1.459
-0.198
-0.462
-0.132
-0.143
0.142
1.749
(0.212)
(1.342)
(0.185)
(1.435)
(0.182)
(2.122)
(0.202)
(1.506)
(0.187
)(1.674)
(0.220)
(1.926)
Parents’SES
NSSEC
1-2
0.015
-0.723*
0.009
-0.187
0.046
0.041
0.128**
-0.582
-0.026
-0.437
0.006
-0.054
(0.070)
(0.398)
(0.062)
(0.387)
(0.059)
(0.543)
(0.063)
(0.411)
(0.061
)(0.418)
(0.063)
(0.480)
NSSEC
30.137
-0.747
0.025
-1.657**
0.185*
-1.277
0.022
-0.593
-0.026
-0.818
0.044
0.649
(0.092)
(0.565)
(0.092)
(0.655)
(0.096)
(0.869)
(0.092)
(0.561)
(0.095
)(0.667)
(0.101)
(0.759)
NSSEC
4(self-
-0.254***
0.127
-0.144*
0.224
-0.053
0.550
0.062
0.977*
0.017
1.305**
0.016
1.533**
employed
(0.083)
(0.458)
(0.084)
(0.580)
(0.097)
(0.783)
(0.081)
(0.502)
(0.091)
(0.601)
(0.095)
(0.729)
NSSEC
8(lon
g--0.039
-0.513
0.078
-1.543
-0.080
-0.070
0.023
-0.308
0.062
-0.159
0.117
0.578
term
unem
ployed)
(0.136)
(0.971)
(0.141)
(1.184)
(0.137)
(1.243)
(0.122)
(1.328)
(0.128)
(1.242)
(0.154)
(1.349)
Lon
eparent
0.035
-0.023
0.066
0.080
-0.022
-0.471
-0.028
0.964*
0.122
0.617
0.172**
1.353**
(0.080)
(0.516)
(0.073)
(0.539)
(0.077)
(0.701)
(0.087)
(0.566)
(0.076
)(0.528)
(0.079)
(0.640)
Observation
s4854
4854
4854
4846
4846
4846
Notes:*:p<
0.1;**:p<
0.05
***:p<
0.01.Sta
ndard
errors,clu
stere
dby
school,
inpare
nth
ese
s.Longitudin
alweights
applied.
63
Tab
leA10
:Com
plete
max
imum
likelihoodestimationresults:
Child-leadingmodel
(continued
)
Boys
Girls
Wave1
Wave2
Wave3
Wave1
Wave2
Wave3
tran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rstran
sfer
WorkHou
rs
You
th’s
ethnicity
Mixed
race
-0.023
-3.032***
0.109
-2.617**
0.074
-2.247**
-0.094
-0.437
-0.023
-0.055
-0.035
-1.346
(0.132)
(0.948)
(0.125)
(1.126)
(0.128)
(0.978)
(0.132)
(0.755)
(0.139)
(0.860)
(0.122)
(1.200)
Indian
-0.520***
-3.602***
-0.410***
-4.029***
-0.326***
-5.698***
-0.619***
-5.734***
-0.573***
-7.132***
-0.360***
-9.613***
(0.096)
(0.685)
(0.093)
(0.964)
(0.102)
(1.068)
(0.112)
(1.216)
(0.105)
(1.045)
(0.112)
(1.222)
Pak
istanior
0.157
-6.248***
0.108
-7.117***
0.301**
-7.575***
0.114
-9.147***
0.115
-13.012***
0.092
-11.396***
Ban
glad
eshi
(0.113)
(0.890)
(0.124)
(1.040)
(0.133)
(1.275)
(0.114)
(1.483)
(0.114)
(2.094)
(0.117)
(1.439)
Black
African
or-0.001
-2.087*
0.069
-2.564*
0.090
-5.208**
0.018
-1.844**
-0.013
-2.882***
0.304**
-3.886***
Caribbean
(0.120)
(1.254)
(0.141)
(1.333)
(0.125)
(2.061)
(0.108)
(0.920)
(0.113)
(1.006)
(0.124)
(1.056)
Other
0.042
0.131
-0.163
-1.934
0.073
-3.706**
-0.342*
-2.447*
-0.507***
-3.038
-0.514***
-5.429***
(0.206)
(1.212)
(0.191)
(1.448)
(0.209)
(1.776)
(0.192)
(1.482)
(0.158)
(1.861)
(0.164)
(1.934)
SpecialEd’Needs
-0.091
-0.411
-0.035
-0.841
-0.065
-0.787
-0.057
-0.477
0.003
-0.993
-0.123
-1.937***
(0.067)
(0.441)
(0.064)
(0.522)
(0.067)
(0.623)
(0.085)
(0.614)
(0.081)
(0.623)
(0.082)
(0.687)
Tow
nor
Village
-0.077
-0.099
-0.164**
1.169***
-0.123*
1.546***
-0.161**
1.263***
-0.251***
2.159***
-0.243***
2.422***
(0.071)
(0.379)
(0.064)
(0.448)
(0.065)
(0.523)
(0.069)
(0.444)
(0.068)
(0.404)
(0.065)
(0.456)
Isolated
area
-0.323**
0.013
-0.279**
1.190
-0.533***
3.036***
-0.082
1.011
-0.245*
3.486***
-0.077
3.287***
(0.146)
(0.724)
(0.122)
(0.892)
(0.124)
(1.080)
(0.127)
(0.836)
(0.130)
(0.721)
(0.146)
(0.859)
Illhealthlimits
0.153**
0.126
0.072
-0.072
0.043
-0.187
-0.048
-0.335
0.030
-0.948**
0.038
-0.131
parent’sactivities
(0.068)
(0.417)
(0.067)
(0.504)
(0.062)
(0.545)
(0.062)
(0.425)
(0.061)
(0.410)
(0.066)
(0.532)
Hom
eow
ner
(mortgage
-0.065
0.047
0.102
0.316
-0.108*
0.221
-0.064
-0.609
-0.058
0.240
0.018
0.128
oroutright)
(0.064)
(0.438)
(0.067)
(0.505)
(0.063)
(0.634)
(0.071)
(0.458)
(0.069)
(0.480)
(0.070)
(0.547)
Stepfamily
0.096
-0.117
0.071
-0.263
-0.008
-0.081
0.135*
0.222
0.015
0.024
-0.091
-0.000
(0.075)
(0.464)
(0.072)
(0.515)
(0.068)
(0.631)
(0.081)
(0.490)
(0.077)
(0.507)
(0.073)
(0.536)
Not
GreaterLon
don
-0.176**
2.158***
-0.094
1.369*
-0.067
2.372***
0.080
2.099***
0.116
2.287***
0.128
0.929
(0.086)
(0.678)
(0.096)
(0.790)
(0.093)
(0.794)
(0.085)
(0.673)
(0.077)
(0.707)
(0.085)
(0.698)
Month
ofinterview:
March
-0.096
0.212
-0.046
0.316
-0.033
-0.329
-0.024
0.421
0.012
1.803**
-0.085
-0.522
(0.064)
(0.371)
(0.129)
(0.865)
(0.056)
(0.420)
(0.069)
(0.438)
(0.128)
(0.865)
(0.054)
(0.402)
April
-0.051
0.373
-0.101
0.497
-0.058
1.020
-0.023
0.309
0.008
2.189**
0.052
1.096
(0.067)
(0.434)
(0.128)
(0.867)
(0.080)
(0.718)
(0.073)
(0.462)
(0.130)
(0.874)
(0.078)
(0.674)
May
-0.054
0.932*
-0.092
0.403
-0.318**
-0.671
-0.041
1.181**
0.062
1.742*
-0.306***
-0.776
(0.086)
(0.524)
(0.131)
(0.924)
(0.139)
(1.260)
(0.091)
(0.535)
(0.131)
(0.896)
(0.118)
(1.098)
June
-0.083
1.168
-0.120
1.564
-0.216
-0.159
-0.020
1.051
-0.087
1.709*
-0.145
3.388*
(0.111)
(0.887)
(0.138)
(0.974)
(0.215)
(2.222)
(0.118)
(0.969)
(0.137)
(0.960)
(0.207)
(1.836)
July
0.194
0.634
0.155
1.242
-0.731
.0.058
0.472
-0.088
2.442
0.496
0.929
(0.151)
(0.862)
(0.202)
(1.349)
(0.744)
.(0.158)
(0.948)
(0.200)
(1.592)
(0.533)
(3.815)
After
Birthday
-0.149***
0.208
-0.012
0.280
-0.009
-0.406
-0.007
-0.015
-0.034
-0.237
-0.042
-0.684
(0.053)
(0.461)
(0.061)
(0.585)
(0.048)
(0.703)
(0.058)
(0.539)
(0.063)
(0.486)
(0.052)
(0.628)
Observations
4854
485
44854
4846
4846
4846
Notes:*:p<
0.1;**:p<
0.05
***:p<
0.01.Sta
ndard
errors,clu
stere
dby
school,
inpare
nth
ese
s.Longitudin
alweights
applied.
64